Cropland concentration powers sustainable intensification of agriculture in China

Abstract

Over the past few decades, China’s cropland configuration has changed markedly, with important implications for food security and agricultural sustainability. Here we examine how cropland spatial-temporal concentration relates to agricultural intensification across China from 2000 to 2020. We mapped cropland distribution for five benchmark years using Google Earth Engine platform and 234,804 Landsat images and quantified cropland landscape configuration. These data were combined with provincial statistics on crop yield, fertilizer application, irrigation water use, and agricultural machinery power. We find that cropland experienced a contraction phase from 2000 to 2015 followed by a recovery phase after 2015. During contraction, increases in crop production commonly coincided with rising fertilizer and machinery inputs, whereas during recovery these inputs declined in most regions, alongside increased cropland contiguity. Overall, agricultural intensification in China became more sustainable by reducing reliance on fertilizer and machinery inputs and placing greater emphasis on cropland-use efficiency and environmental protection.

Introduction

Food security is a fundamental guarantee for human survival and development¹. Despite a doubling of global crop production over the past half-century, global food security still faces severe challenges². By the middle of this century, the global population is projected to reach 9 billion, with crop demand increasing by 100–110%³. To address this trend, agricultural intensification, which aims to enhance the productivity of existing cropland, has been widely recognized as a key strategy to alleviate pressure on food security.

However, traditional agricultural intensification primarily focuses on maximizing yield, often relying on the extensive use of fertilizers, mechanization, and pesticides⁴. While such practices can boost crop production, they also pose serious environmental threats, including non-point source pollution, greenhouse gas emissions, and soil degradation⁵. In China, approximately 30% of global fertilizer and pesticide consumption is concentrated¹⁰. Although this contributes to relatively high grain production, it has also led to the substantial loss of high-quality cropland⁶.

China has long faced structural challenges such as land fragmentation, rural population aging, and rapid urban expansion^7,8. Under the framework of the household responsibility system, combined with underdeveloped land transfer mechanisms, farmland has remained highly fragmented and small-scale for decades^9,10. Smallholder farmers are often associated with excessive use of agrochemicals, low labor efficiency, and uneven resource allocation¹¹, further constraining the improvement of agricultural intensification.

Against this backdrop, sustainable agricultural intensification has emerged as a key agenda for coordinating agricultural development and environmental protection. The early definition of sustainable intensification stated: “increase yield without expanding cropland area or causing adverse environmental effects¹².” In contrast to traditional intensification, the current approach emphasizes achieving higher yields with the same or fewer inputs while simultaneously ensuring food security and environmental sustainability^13,14. Although genetically modified technologies offer the potential to increase yield without increasing environmental pressure, public resistance persists in some regions due to policy and ethical concerns^15,16. Therefore, centralized cropland management, as a more practical and socially acceptable alternative, has gained increasing attention in both academic and policy circles¹⁷. Its potential lies in improving resource allocation efficiency and thereby enhancing yield per unit area¹⁸.

The United Nations has incorporated indicators to assess sustainable agriculture into its Sustainable Development Goals (SDG), notably SDG 2.4.1¹⁹. Yet, such indicators primarily rely on household surveys, government statistics, and other ground-based data²⁰, limiting their applicability in large-scale and efficient assessments. In recent years, researchers have begun to explore the use of remote sensing technologies to evaluate the sustainability of agricultural intensification.

Although multiple global or regional land cover products have been developed^21,22,23,(e.g., GLC500m, MCD12Q1, GlobeLand30), these products focus on general land cover classification rather than the identification of actively cultivated cropland, resulting in cropland classification accuracy varying between 60 and 80%^24,25. Moreover, most existing studies focus on static time points, lacking large-scale and long-term dynamic evaluations of sustainable agricultural intensification in China²⁶. In addition, although cropland concentration is widely considered an effective strategy to promote sustainable intensification¹⁷, there is still a lack of systematic evidence and broad consensus regarding its actual impact in the Chinese context.

We developed a systematic evaluation framework for sustainable agricultural intensification in China. Based on the conceptual foundation of sustainable intensification, we integrated remote sensing and machine learning techniques to extract actively cultivated cropland data at 30-m resolution for five representative years from 2000 to 2020. Spatial changes in cropland patterns are assessed using three landscape metrics, including mean patch area (MPA), patch density (PD), and aggregation index (AI)²⁷. In parallel, we incorporate national statistical data on fertilizer input, machinery power, irrigation water use, and crop yield to evaluate the sustainability of agricultural intensification. Finally, we compare cropland pattern data with farm size data to assess China’s potential for large-scale cropland management and the future development of sustainable intensification.

Results and discussion

Cropland data

The overall accuracy of data extraction from the 5 years is above 93% (Supplementary Table 1; 93.78% for 2000, 95.18% for 2005, 95.53% for 2010, 94.04% for 2015, and 95.18% for 2020, respectively). This accuracy is greater compared to other results^24,25. Linear regression model was used to correlate remote sensing-derived data with the cropland area extracted from statistical yearbook data, which yielded high accuracy with an R² of 0.85 (Fig. 1b, P < 0.01; Classification accuracy is listed in Supplementary Table 1). The remote sensing-based extraction results for China’s cropland in 2015 were consistent with the previous study²⁸, and the overall trend of cropland area changes is also similar to that of the related study²⁹. These together validate the reliability of our results.

Fig. 1: Evaluation method and extraction results.

a Changes in agricultural production indicators from 2000 to 2020. b Comparison of extracted cropland area with cropland area from national statistics data. c Changes in cropland area from 2000 to 2020. d Evaluation matrix of change patterns on the contribution to sustainable agricultural intensification.

Spatial-temporal distribution characteristics of cropland were analyzed (Supplementary Fig. 1). The average cropland area over these five years was 178.23 ± 5.56 Mha, representing 18.57% of China’s land area. Cropland distribution in China shows a clear geographical variation. The majority of cropland is concentrated in the eastern regions (mainly in NCP, HHHP, MLYP and SC). Cropland in these regions is characterized by hills and plains with small topographic variations. Rainfall and sunshine are abundant, conducive to development of concentrated cropland. The western regions (including NASR, LP, SCB, QTP, and YGP) account for 72.11% of the country’s land area, but only contribute 36.72% of the average cropland area from 2000 to 2020. These regions are mainly characterized by large mountains, highlands, and basins with large topographic variations. Rainfall is relatively less and the cropland is mostly fragmented.

Cropland area at 2015 is the lowest with only 169.26 Mha. Therefore, the year 2015 is considered as the turning point and the study period was divided into two phases: the contraction period (2000–2015) and the recovery period (2015–2020) (Fig. 1c). At the contraction period, the total cropland area in China decreased 14.87 Mha. This phenomenon can be attributed to the accelerated urbanization and industrialization at this period, which led to the transformation of cropland into more economically profitable land-use types³⁰. From 2000 to 2015, China’s urbanization rate grew from 36.2 to 56.1%. The accelerated process of urbanization increased the demand for rural labor to migrate to cities. This leads to a reduction of 204.91 million rural populations and a lack of agricultural labor force, and some cropland becomes desolate or abandoned^31,32. Furthermore, construction of large projects, such as water conservancy facilities, natural vegetation protection and ecological conservation projects, highways, high-speed railways, etc., led to a contraction of the cropland area. For instance, the construction of the Three Gorges Reservoir leads to a 4% decrease in the cropland area of the reservoir³³. Moreover, since 1998, the Chinese government has implemented a series of policies and plans to restore ecosystems and improve environmental protection. One of these is the “Grain for Green” program, which requires transforming cropland unsuitable for farming into other land-use types, such as forests, shrubs, and grasslands. Data released by China’s National Forestry and Grassland Administration shows that by the end of 2015, the “Grain for Green” program had reduced the cropland area by 26.7 Mha³⁴.

After 2015, the change of cropland area entered a recovery period. In 2020, the total area of cropland in China was 177.18 Mha. Compared to 2000, the total cropland area in 2020 decreased 6.34 Mha. However, there was an increase of 8.52 Mha compared with 2015. In recent years, the Chinese government has implemented numerous measures to protect cropland. For instance, the enactment of the farm circulation policy encouraged rural collective economic organizations and farmers to transfer land to modern agricultural enterprises for large-scale operations³⁵. This, in turn, promoted the intensive use of land and led to an increase in cropland area. The land occupation compensation balance was implemented to strike a balance between local development and land protection³⁶. Furthermore, beginning in 2006, the Chinese government strongly emphasized national food security. China implemented a government-led responsibility mechanism for cropland protection to strictly uphold the “red line” of 1.8 billion mu (ca. 12 Mha) of cropland. The construction of high-standard farmland is vigorously promoted following the order of cropland and permanent basic farm, ecological protection redlines, and urban development boundaries³⁷. These measures optimize the structure of cropland and contribute to the achievement of sustainable agricultural development.

Changes in agricultural production indicators

Changes in agricultural production indicators show that from 2000 to 2020, the yield of major food crops in China increased continuously. Meanwhile, the degree of agricultural intensification was improved during this period, which is of great significance for ensuring national food security and self-sufficiency. In terms of agricultural inputs, this study respectively analyzed fertilizer use, agricultural machinery power, and irrigation water use (Fig. 1a). From 2000 to 2020, fertilizer use increased to 360 kg ha⁻¹ by 2015 and thereafter decreased to 303 kg ha⁻¹ by 2020. This indicates that at the contraction period, growth of crop yield relied strongly on the extensive use of fertilizers. The decrease from 2015–2020 can be attributed to the government’s efforts in recent years to reduce pollution caused by excessive use of fertilizers, including implementation of a series of measures (e.g., cancellation of fertilizer subsidies, soil testing and optimization of chemical fertilization use, vigorous promotion of organic fertilizer use, and zero-growth actions for fertilizers and pesticides)³⁸. Similarly, machinery power usage also showed an increase from 2000–2015 and then dropped to 6.31 kW ha⁻¹ by 2020. On the one hand, the Chinese government has undertaken lots of reforms to tackle agricultural pollution. In 2015, the Chinese government established the goals and measures to prevent and control agricultural non-point source pollution³⁹. On the other hand, the rural population dropped by approximately 80 million people from 2015 to 2020, and the level of agricultural mechanization approached saturation. This resulted in a decrease in the total agricultural machinery power usage in China and a decline in demand of the agricultural machinery market^40,41. China’s agricultural machinery investment is developing toward intelligence, efficiency, and environmental friendliness⁴². For instance, from 2015 to 2020, the market size of China’s plant protection drones increased from 1.62 billion yuan to 6.79 billion yuan, and the market size increased to 13.1 billion yuan in 2021⁴³. In contrast, driven by policy interventions^44,45, irrigation water use per unit area in China has continuously declined over the past two decades, indicating reduced irrigation input and improved water use efficiency, thereby promoting sustainable agricultural intensification.

The relationship between agricultural inputs and outputs in the nine agricultural production zones from 2000 to 2020 was analyzed to understand the changing trends of agricultural production indicators in China more comprehensively (Fig. 2). The trends of fertilizer use in each agricultural production zone were consistent with the national level. All zones exhibited an increase from 2000 to 2015 and then declined until 2020, reflecting the strict national implementation of the zero-growth action for fertilizer use launched in 2015. Despite a decrease in fertilizer use in the zones of NASR, LP, MLYP, SCB, YGP, and SC at the recovery period, crop yield increased. This suggests that the role of fertilizer use in regulating agricultural intensification is weakened.

Fig. 2: Relationship between agricultural production indicators and crop yield in nine agricultural production zones from 2000 to 2020.

Note: NCP (Northeast China Plain), NASR (Northern arid and semiarid regions), HHHP (Huang-Huai-Hai Plain), LP (Loess Plateau), QTP (Qinghai Tibet Plateau), MLYP (Middle-Lower Yangtze Plain), SCB (Sichuan Basin and surrounding regions), YGP (Yunnan-Guizhou Plateau), SC (Southern China).

From 2000 to 2020, the total agricultural machinery power usage in China’s nine agricultural production zones all increased, indicating an overall rise in the level of agricultural mechanization. Only the MLYP, SCB, and SC zones saw a yearly increase in machinery power usage, indicating that these zones have achieved large progress in advancing agricultural mechanization. In contrast, after a clear increase in total agricultural machinery power usage at the contraction period, other zones note a decrease from 2015 to 2020, which may be attributed to more scientific farm management and a reduction of unnecessary agricultural machinery power inputs.

In addition, in the QTP, due to the special geographical environment and climate conditions, the scale of agriculture is limited and crop yield changes greatly. The relationship between crop yield in this area and total agricultural machinery power usage and fertilizer use is relatively complex, presenting characteristics different from other zones.

Current status of sustainable intensification of agriculture in China

The number of regions for each combination of change patterns was counted (Supplementary Table 2) and visualized to exhibit the degree and trends of sustainable intensification of agriculture in China (Fig. 3). From a temporal perspective, agricultural intensification at the contraction period (2000–2015) was primarily realized by high agricultural input (e.g., machinery power usage, chemical fertilizer use). Most regions are classified as “Negative” and “Highly negative” patterns. The M-I and F-I factors contribute to the prevalence of “Negative” and “Highly negative” pattern combinations in most regions. Out of the 31 regions, 24 regions experienced an increase in fertilizer use, and 30 regions experienced an increase in agricultural machinery power usage, which coincided with increased crop yield in 24 regions. At this period, the shape of cropland became more fragmented (MPA-D, PD-I, AI-D). It should be noted that some regions were classified as “Positive” and “Neutral” patterns solely due to an increase in crop yield (Y-I). This observation suggests that the yield increase during this period was closely associated with rising fertilizer application, alongside concurrent increases in machinery power usage (Supplementary Fig. 2)⁴⁶. During this period, although only three regions experienced an increase in irrigation water use per unit area (W-I), most regions still exhibited increasing cropland fragmentation, suggesting that improvements in irrigation water efficiency alone had a limited effect on promoting sustainable agricultural intensification. China experienced rapid agricultural intensification during this period, but unfortunately, this phenomenon is not sustainable⁴⁷.

Fig. 3: Number of regions for each combination of change patterns.

a the contraction period from 2000 to 2015. b The recovery period from 2015 to 2020. The X and Y axes, respectively, reflect the change patterns of landscape indexes and agricultural production indicators, which together determine the pattern combination of a region. The height of the Z axis represents the number of regions under each combination. Color mapping is used to represent the contribution level of these combinations to sustainable agricultural intensification.

At the recovery period (2015–2020), factors controlling agricultural intensification shift from the previous heavy dependence on high inputs of fertilizer and machinery power to a more balanced approach involving cropland concentration and a reduction in the use of fertilizer and machinery power. Meanwhile, irrigation water use efficiency continued to improve, with irrigation water use per unit area showing a sustained decline, further reinforcing the dual trend of reducing agricultural input intensity and improving resource use efficiency. Cropland area increased alongside cropland concentration, primarily characterized by three change patterns: MPA-I, PD-D, and AI-I. The number of regions belonging to these patterns is 4 at the contraction period, but raised to 64 at the recovery period (some regions have multiple change patterns), suggesting the reinforcement of cropland concentration at the recovery period. The number of regions with F-I and M-I patterns reduced to 1 and 4, respectively, in accordance with our previous finding that fertilizer use and machinery power usage are weakened. Furthermore, the number of regions with “Highly positive” and “Positive” patterns increased. This indicates that in most regions, the approach to increasing yields has shifted from the previous heavy reliance on high inputs of fertilizer and machinery power to a combination of cropland concentration and reduced inputs of fertilizer and machinery power. This shift suggests that agricultural intensification in China becomes more sustainable. This also reflects the evolution of China’s farmer fertilizer regulation policy from “increasing quantity and yield” to “reducing quantity and increasing efficiency”. Lots of policies and measures, including promotion of organic fertilizer use, improvement of soil quality, implementation of crop rotation and fallow system, are promoted^38,48,49,50.

The change patterns of the 31 regions were used to assess the extent of sustainable intensification across China (Fig. 4, Supplementary Table 3). During the contraction period, except for Xinjiang, Ningxia, and Jiangxi, all other regions were graded as D or below, indicating that agricultural intensification in most provinces is unsustainable. In contrast, at the recovery period, except for Hainan which was below Grade D, other regions were all graded as or above C. Particularly, 16 regions were graded as Grade A, including one region classified as Grade A + . Overall, the number of regions with unsustainable intensification of agriculture (i.e., Grade D, E and F) decreased from 28 at the contraction period to 1 at the recovery period.

Fig. 4: The degree of sustainable intensification of agriculture in provincial administrative regions.

a The contraction period from 2000 to 2015. b In the recovery period from 2015 to 2020.

To further verify whether cropland concentration contributes to improved resource use efficiency, we conducted a correlation analysis between cropland landscape configuration indexes and agricultural input intensity indicators across 31 provincial-level administrative units (Fig. 5, Supplementary Fig. 3, Supplementary Table 4). By comparing paired provincial samples between the contraction period (2000–2015) and the recovery period (2015–2020), we observed notable temporal differences in the relationship between landscape patterns and agricultural inputs. Specifically, during the contraction period, larger patch sizes and higher aggregation levels were positively correlated with greater fertilizer and mechanization input intensity, indicating that cropland concentration at that stage did not lead to enhanced resource use efficiency (in fact, widespread cropland fragmentation was observed during this period). In contrast, this relationship was reversed during the recovery period—provinces with more concentrated cropland exhibited lower fertilizer and mechanization intensity per unit yield, suggesting that improvements in landscape configuration were accompanied by enhanced agricultural input efficiency.

Fig. 5: Relationships between cropland landscape configuration and agricultural input intensities.

MPA, PD, and AI represent Mean Patch Area, Patch Density, and Aggregation Index, respectively. The scatter plots illustrate the relationships between landscape metrics (MPA, PD, and AI) and Fertilizer Intensity (t t⁻¹), Mechanization Intensity (kW t⁻¹), and Irrigation Intensity (m³ t⁻¹). Red (contraction period, 2000–2015) and green (recovery period, 2015–2020) data points and dashed lines indicate the provincial-level observations (31 provinces across different years) and corresponding linear regression fits. Shaded bands around the regression lines indicate the 95% confidence intervals.

In comparison, irrigation input intensity showed a relatively stable negative correlation with cropland concentration in both periods, indicating that improvements in irrigation efficiency were likely driven more by water-saving policies and technological progress rather than changes in spatial cropland configuration. These findings further suggest that during the recovery period, cropland consolidation likely contributed to reduced reliance on fertilizers and agricultural machinery, thereby improving input efficiency.

In conclusion, advancing sustainable agricultural intensification in China can be achieved through promoting cropland connectivity, expanding cropland size, improving resource use efficiency, and minimizing agricultural chemical pollution. Meanwhile, the Chinese government has also identified large-scale cropland operation as a central strategy for promoting sustainable intensification, which is expected to effectively alleviate farmland abandonment caused by population aging and labor shortages¹¹.

These findings highlight the distinct pathway of sustainable agricultural intensification in China compared to many developed countries. In Europe and the United States, intensification has largely built upon a pre-existing foundation of large-scale farms, with its main driving force being technological innovation—such as precision agriculture, conservation tillage, and digital farm management systems⁵¹. These approaches focus on enhancing efficiency and reducing environmental impacts, representing a typical “technology-driven” model of intensification.

By contrast, China tends to prioritize land consolidation before promoting technological adoption. Historically, China’s agricultural system has been dominated by highly fragmented smallholder farming⁵². In this context, institutional and policy support for cropland consolidation has played a central role in driving sustainable agricultural intensification. Unlike the technology-led strategies in developed countries, China’s policy-oriented approach offers a distinctive model that may provide valuable insights for other developing countries facing similar smallholder-related challenges.

Potential and challenges for sustainable intensification of agriculture in China

The discrepancy between cropland size and farm size—where farm size refers to the operated land area under unified management (including contracted land plus transferred-in land)—highlights the potential for scaling up agricultural operations through management integration even under fragmented field conditions⁵³. The data of farm size revealed that over 80% of the cropland belongs to small-scale farms (<0.6 ha), and large-scale farms (>16 ha) only account for 10% (Supplementary Fig. 4). This predominance of small-scale farms in China is primarily due to the Household Responsibility System¹⁰. From 2000 to 2020, proportions of croplands larger than 16 ha show an increase and are all greater than 85% of the China’s total (Supplementary Fig. 4). This suggests a substantial potential for the transformation of large-scale cropland into large-scale farm. Particularly, large-scale croplands in the zones of NCP, HHHP, and MLYP account for over 90% (Supplementary Fig. 5; Supplementary Table 5). These areas are traditionally major commodity grain bases in China. The HHHP and MLYP maintained a relatively stable proportion of large-scale cropland from 2000–2020, suggesting that large-scale cropland concentration is conducive to management and operation. This can effectively prevent cropland being abandoned or occupied for non-agricultural purposes, securing China’s agricultural production. Amid the rapid economic and industrial development, China’s fertilizer use intensity and greenhouse gas emissions are greater than those of most other major crop-producing countries⁵⁴. Although some measures have already been implemented, and fertilizer use is declining, considerable potential to improve its use efficiency still exists. Additionally, the Chinese government has been consistently promoting agricultural water pricing reforms and water-saving irrigation⁵⁵, with a special focus on strengthening the ongoing water-saving renovation projects in large- and medium-sized irrigation districts to improve irrigation conditions⁵⁶. Over the past two decades, irrigation water usage per unit area has steadily decreased by approximately 25% (Supplementary Fig. 6a), thanks to the widespread adoption of water-saving irrigation and industrial water recycling technologies. The implementation of these technologies has enabled China to save an average of approximately 420 billion m³ of water annually, nearly equivalent to 80% of the actual water usage during that period. This contributes to improved water-use efficiency and a more optimized water-use structure⁵⁷. Furthermore, with the strategic implementation of a series of government policies and the formulation of medium to long-term plans^58,59,60,61, China is determined to elevate the efficiency of cropland management and preserve the agricultural ecological environment. This will further unleash the potential for sustainable intensification of agriculture in China.

Despite the great potential of sustainable intensification of agriculture in China, some challenges may exist to influence its development. Climate change has exerted a clear impact on China’s water resources and agriculture. The shifts in precipitation patterns and the escalation of extreme weather events pose a substantial threat to food production. Over the last 50 years, rising temperatures and intensifying droughts in Eastern China have adversely affected crop yields. Although Northern China encompasses 65% of the cultivable land, it possesses a mere 18% of the total water resources, while the Southern part grapples with flooding issues, creating hurdles for sustainable agriculture⁶². The variations in the types of major crops give rise to substantial disparities in irrigation water usage across different agricultural production zones (Supplementary Fig. 6b–j), thereby posing challenges for the efficient utilization of water resources. However, on a nationwide scale, the impact of climate change on agriculture remains limited, owing to the relatively mild nature of climate shifts and the evident positive influence of modern agricultural practices⁶³. Besides, under the broad context of climate change, the water requirements for crops in China might decrease in the future due to the influence of the increased CO₂ levels on plant growth⁶⁴. China has also consistently prioritized enhancing the storage and management of water resources. Projects such as the South-to-North Water Diversion Project play a pivotal role in managing water resource distribution, mitigating flood risks, and alleviating the impacts of droughts, thus bolstering sustainable agriculture⁶⁵.

In addition to water resource management, optimizing the spatial configuration of cropland is also a crucial pathway toward achieving sustainable intensification. Such optimization is not merely a structural adjustment, but a spatial management strategy that balances productivity, resource efficiency, and environmental protection. It offers a potential solution to the bottlenecks in transitioning from conventional to sustainable agricultural intensification⁶⁶. Spatial optimization of cropland can amplify the effectiveness of existing national initiatives—such as soil testing-based fertilization, high-standard farmland construction, and irrigation reform—by enabling their efficient implementation across larger, contiguous land units. For example, measures such as incentives for land transfer, ecological compensation in consolidated areas, and cross-regional coordination mechanisms for cropland can simultaneously enhance productivity and promote ecological restoration⁶⁷.

The results of the 2040 baseline simulation indicate that provincial cropland configurations remain largely stable, with only slight changes in patch size, fragmentation, and aggregation. Future cropland changes may be dominated by stability rather than dramatic reorganization, and China’s cropland landscapes are more likely to undergo gradual structural adjustment (Supplementary Note 7).

While encouraging the scaling-up of farm operations and sustainable intensification of agriculture are emphasized, the potential negative impacts are also worth considering⁶⁸. Socio-economically, these impacts include the reduction of land tenure security, decreased self-sufficient production and the fading of cultural customs⁶⁹. From an ecological and environmental perspective, sustainable intensification of agriculture might lead to biodiversity loss, alterations in the structure of agroecosystems, and a reduction in soil functionality^70,71,72. Moreover, disseminating such concepts of sustainable intensification to millions of smallholder farmers in China is a long-term and meticulous task, as it may affect their livelihoods⁷³. Therefore, when formulating relevant policies in the future, timely consideration and the adoption of reasonable measures to prevent the occurrence of the above-mentioned issues are necessary.

Final remarks

The sustainable transformation of agricultural intensification in China fundamentally depends on the spatial optimization of cropland patterns. As cropland becomes increasingly consolidated and contiguous, the reliance on high-intensity inputs such as chemical fertilizers and machinery tends to decrease, resulting in improved resource use efficiency. We provide theoretical justification and empirical evidence supporting the hypothesis that cropland consolidation is a key driver of the shift from unsustainable to sustainable intensification in China. These findings offer concrete proof of China’s progress toward sustainable agricultural intensification and may serve as a valuable reference for other smallholder-dominated countries and regions pursuing similar pathways.

Materials and methods

Data sources

Landsat series from 1999–2021 in the Google Earth Engine (GEE) platform were used, including apparent reflectance data of Landsat 5, Landsat 7, and Landsat 8. A three-year image set, including the target year, the year before, and the year after, was selected for each target year to represent cropland conditions and address limited data availability in some regions. A total of 234,804 images were extracted (Supplementary Table 6). Due to the differences in spectral reflectance from previous sensors, conversion coefficients were used to normalize the OLI reflectance to the reflectance of TM and ETM+ (Supplementary Table 7)⁷⁴. In addition, Sentinel-2 images from the GEE platform, and the Digital Elevation Model (DEM) terrain dataset derived from the Shuttle Radar Topography Mission (SRTM) were also included. Farm size data in China was derived from the Global Distribution of Farm Size Dataset for 2017 (http://pure.iiasa.ac.at/id/eprint/15526/)⁷⁵. Administrative division data, including national administrative boundary vector data and vector data for the nine agricultural production zones, were all acquired from the Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences (RESDC, https://www.resdc.cn/). Provincial agricultural irrigation water usage data were obtained from the Water Resources Bulletins publicly available on the official website of the Ministry of Water Resources of the People’s Republic of China (http://www.mwr.gov.cn).

The agricultural production indicators are derived from the China Statistical Yearbooks released at 2001, 2006, 2011, 2016, and 2021 (http://www.stats.gov.cn/tjsj/ndsj/), respectively. These indicators include the amount of fertilizer use (100 kg ha⁻¹, abbreviated as F), total machinery power usage (kW ha⁻¹, abbreviated as M), and total crop yield (t ha⁻¹, abbreviated as Y) at the provincial level, the detailed information is provided in Supplementary Note 1.

In the process of data selection and the calculation of cropland fragmentation metrics, potential uncertainties may arise from both remote sensing errors and socioeconomic data fluctuations. We prioritized the use of Landsat series imagery as the primary data source due to its long temporal span (consistently available since the 1980s), moderate spatial resolution (30 m, suitable for identifying cropland patches at the regional scale), public accessibility, and well-established preprocessing workflows.

Although Sentinel-2 offers higher spatial resolution (10 m), its relatively short temporal coverage (available only since 2015) limits its utility for long-term change detection beginning in 2000. MODIS imagery provides long-term and high temporal resolution data, but its coarse spatial resolution (250–1000 m) makes it unsuitable for capturing small and fragmented cropland parcels, which are essential for landscape pattern analysis. Therefore, Landsat represents a balanced choice in terms of spatial and temporal resolution.

Nevertheless, some uncertainties persist during classification and metric computation, including cloud shadow interference, inter-sensor inconsistencies, and vegetation classification errors. To reduce these uncertainties and enhance data consistency, we selected five discrete target years (2000, 2005, 2010, 2015, and 2020) and adopted a three-year composite strategy centered on each target year. While this approach improves classification accuracy and cross-year comparability, it may fail to capture abrupt changes in cropland patterns caused by natural disasters, extreme weather events, or intensive regional development.

In addition, although the China Statistical Yearbook provides government-reported data with generally high reliability, it may still contain missing values or reporting errors at certain time points. To improve consistency across data sources, we cross-validated the statistical records with our remote sensing results to enhance interpretability and data robustness.

A quantitative uncertainty analysis was performed to evaluate both the potential errors in the input data and their propagation through the computational workflow, providing insights into how these variations may affect the reliability and robustness of the study’s conclusions (Supplementary Note 8; Supplementary Fig. 9).

Methods

Extraction of cropland distribution data based on Random Forest

The time series viewer was developed using the GEE platform to enable sample selection⁷⁶. The Bare Soil Index (BSI), Normalized Difference Vegetation Index (NDVI), and Brightness, Greenness, Wetness (Supplementary Table 8) obtained through Tasseled Cap transformation for the years 2000, 2005, 2010, 2015, and 2020 were calculated. Sentinel-2 imagery (10 m) and corresponding cloud-free Landsat 5–8 and Sentinel 2 false-color (RGB: NIR, SWIR1, Red) composite images were then combined with these indexes to enable the stable identification of cropland and non-cropland (Supplementary Fig. 7). In our study, we took the lands that are being used for growing crops as cropland samples. Based on this viewer, visual inspection and manual selection of cropland and non-cropland pixels that remained relatively stable throughout the study period were performed. The principles of uniformity and randomness were followed to select samples. In total, 2270 cropland samples and 3431 non-cropland samples were identified (Fig. 6a). These samples were randomly divided into training and validating datasets, respectively, based on a ratio of 8:2, the validation set and the training set are independent of each other⁷⁷. The data were then processed with ArcGIS and uploaded to GEE Assets for further analysis.

Fig. 6: Spatial distribution of samples and feature variables used in the Random Forest classifier.

Note: B1, B2, B3, B4, B5, and B7 are the spectral reflectance data of the Landsat blue, green, red, NIR, SWIR1, and SWIR2 bands; NDVI is the Normalized Difference Vegetation Index; B1_P75, B1_P75_1, B2_P75, B2_P75_1, B3_P75, B3_P75_1, B4_P75, B4_P75_1, B5_P75, B5_P75_1, B7_P75, B7_P75_1, NDVI_P75, and NDVI_P75_1 are the 75th percentiles of the blue, green, red, NIR, SWIR1, and SWIR2 bands and NDVI for every two years, respectively; NDWI is the Normalized Difference Water Index; diff, min, and stddev are the range, minimum value, and standard deviation of NDVI, respectively. Slope was derived from the SRTM DEM.

Spectral features, vegetation indexes, and terrain features from Landsat remote sensing data were used to create feature variables in a Random Forest Classifier. Totally, 44 variables were included as input feature variables for the Random Forest Classifier (Supplementary Table 9). The differences of each variable between cropland and non-cropland samples were illustrated in Fig. 6b, where NDVI and Normalized Difference Water Index (NDWI) show greater discriminability between cropland and non-cropland samples across all regions. The separability between cropland and non-cropland is the highest for the 2nd and 3rd periods, both of which are the peak growing seasons for crops in China. Similarly, the maximum separability between cropland and pasture occurs in the 3rd and 6th periods. The presence of apparent outliers is mainly induced by the noise among the data of several band values (detailed information is provided in Supplementary Note 2).

The boundary vector data and all training samples were imported into GEE via GEE Assets to extract cropland distribution data. Validating dataset was used to evaluate the accuracy of the distribution data. A first-degree polynomial fitting was used to assess its reliability, and the changes of cropland size with time was correlated using a fourth-degree polynomial regression. Additionally, since the overarching agricultural policy in China is set by the central government and implemented by provincial agricultural authorities, an assessment based on provincial administrative regions can provide a representative evaluation⁷⁸. This is also a common approach for analyzing regional agricultural production in China^79,80,81. Therefore, we performed a projection transformation of the extraction results in ArcGIS and calculated the cropland area of 31 regions (including all provinces, municipalities directly under the Central Government, and autonomous regions in China) for the years 2000, 2005, 2010, 2015, and 2020. These results were compared with cropland area derived from the national statistical data. Due to data omission, Hong Kong, Macao, and Taiwan regions are not included in this study (Supplementary Note 3 provides the specific method for accuracy evaluation).

Regional division

The influence of climate, terrain, and other factors results in distinctive regional characteristics of crop cultivation in China. We used the division of nine agricultural production zones to further analyze the relationships between landscape indexes and agricultural production indicators (Supplementary Fig. 8). These nine agricultural production zones are Northeast China Plain (NCP), Northern arid and semiarid regions (NASR), Huang-Huai-Hai Plain (HHHP), Loess Plateau (LP), Qinghai-Tibet Plateau (QTP), Middle-Lower Yangtze Plain (MLYP), Sichuan Basin and surrounding regions (SCB), Yunnan-Guizhou Plateau (YGP), and Southern China (SC), respectively⁸². Hong Kong, Macao, and Taiwan were excluded from this study due to data limitations in these regions (A detailed description of the production areas is provided in Supplementary Note 4).

Division of change patterns

For the division of change patterns of landscape indexes and agricultural production indicators, we introduced the Freedman-Diaconis rule⁸³, with the bin width set as follows:

$$H=2\times {{{\rm{IQR}}}}\times {N}^{-\frac{1}{3}}$$

(1-1)

Where H is the width, IQR is the interquartile range of the data, and N is the number of provincial administrative regions studied and equals to 31. Methods for calculating the rate of change in landscape indexes and agricultural production indicators are provided in Supplementary Notes 5 and 6. The Freedman–Diaconis rule was originally developed for probability density estimation and histogram binning. In this study, it was not used to directly model the temporal dynamics of agricultural or landscape indexes, but rather served as a statistical basis for dividing indicator change intervals, in order to enhance the objectivity and robustness of the binning criteria.

We define scenarios where the degree of change in landscape indexes and agricultural production indicators is greater than “H” as an increase (“I”) pattern, including MPA-I, PD-I, AI-I, F-I, Y-I, W-I and M-I. In contrast, when the change is less than “-H”, it is defined as a decrease (“D”) pattern, including MPA-D, PD-D, AI-D, F-D, Y-D, W-D and M-D. When the degree of change falls between “-H” and “H”, it is defined as a stable (“S”) pattern, including MPA-S, PD-S, AI-S, F-S, Y-S, W-S and M-S (detailed division of the change patterns is provided in Supplementary Table 10).

F-I, M-I, W-I and Y-D patterns, deemed as “negative”, represent increased dependency on fertilizer use, machinery power, and irrigation water, along with a decline in crop yield. These patterns indicate higher environmental pressure and unsustainable agricultural intensification. In contrast, F-D, M-D, W-D and Y-I patterns are considered “positive”, as they indicate reduced dependence on fertilizer, machinery, and irrigation water, along with improved crop yield, thereby reflecting lower environmental pressure and a trend toward sustainable agricultural intensification. For landscape indexes, MPA-I, PD-D, and AI-I suggest stronger cropland concentration, which represents sustainable intensification of agriculture. In comparison, MPA-D, PD-I, and AI-D represent cropland fragmentation of cropland, which are indication of unsustainable intensification of agriculture.

Combination of change patterns

A series of combinations of change patterns was formed in each region. These combinations were qualified to evaluate their contribution to the sustainable intensification of agriculture in China. Firstly, we scored each change pattern according to their degree of change. We assigned a score of −1 for change patterns that have a negative impact on cropland change and the environment (i.e., MPA-D, PD-I, AI-D, F-I, M-I, W-I and Y-D). A score of 0 was assigned for change patterns that basically have no impact (i.e., MPA-S, PD-S, AI-S, F-S, M-S, W-S and Y-S). We assigned a score of +1 for change patterns that have a positive impact on cropland change and the environment (i.e., MPA-I, PD-D, AI-I, F-D, M-D, W-D and Y-I). Secondly, an evaluation matrix was established to group these pattern combinations into five categories, namely (Fig. 1d): Highly positive (+2 points), Positive (+1 points), Neutral (0), Negative (−1 point), and Highly negative (−2 points). Each combination corresponds to a score. This enables the qualification of the contribution of these combinations to the sustainable intensification of agriculture. Thirdly, the combination of change patterns for each region was visualized to illustrate the degree and trend of sustainable intensification of agriculture in China.

Development of agricultural input intensity indicators

To analyze the period-specific differences in agricultural resource use associated with changes in cropland landscape configuration, this study used 31 provincial administrative units across China as analytical units. Three agricultural input intensity indicators—fertilizer, mechanization, and irrigation—were developed to represent the use levels of major agricultural resources under comparable unit yield conditions. In parallel, three landscape pattern indexes, including MPA, AI and PD were extracted to characterize the spatial configuration of cropland across provinces. Based on the division into a contraction period (2000–2015) and a recovery period (2015–2020), province-year paired samples were established by matching each province’s landscape configuration with corresponding resource input intensities. Linear regression model were applied to examine the association and changing patterns between cropland landscape structure and agricultural resource use across different periods, to determine whether improvements in landscape configuration are accompanied by enhanced resource use efficiency in agriculture.

The definitions and calculation methods of the three input intensity indicators are described as follows:

(1) Fertilizer Intensity

Fertilizer Intensity measures the amount of fertilizer applied per unit of grain yield. It is defined as the ratio between fertilizer application and crop yield per hectare, with units of tons per ton (t t⁻¹):

$${{{\rm{Fertilizer\; Intensity}}}}=\frac{{\mbox{Fertilizer}}({{{{\rm{t}}}}{{{\rm{ha}}}}}^{-1})}{{\mbox{Yield}}({\mbox{t}}{{{{\rm{ha}}}}}^{-1})}$$

(1-2)

where Fertilizer represents the fertilizer amount applied per hectare (t ha⁻¹), and Yield refers to grain yield per hectare (t ha⁻¹). The ratio thus reflects the quantity of fertilizer required to produce one ton of grain.

(2) Mechanization Intensity

Mechanization Intensity reflects the agricultural machinery power required per unit of output. It is calculated as the ratio between total machinery power per hectare and crop yield, with units of kilowatts per ton (kW t⁻¹):

$${{{\rm{Mechanization\; Intensity}}}}=\frac{{\mbox{Machinery power usage}}({\mbox{kW}}{{{{\rm{ha}}}}}^{-1})}{{\mbox{Yield}}({\mbox{t}}{{{{\rm{ha}}}}}^{-1})}$$

(1-3)

where Machinery power usage is the total agricultural machinery capacity per hectare (kW ha⁻¹), and Yield is the grain output per hectare.

(3) Irrigation Intensity

Irrigation Intensity measures the amount of irrigation water used per ton of crop yield. It is defined as the ratio between irrigation water input and yield per hectare, with units of cubic meters per ton (m³ t⁻¹):

$${{{\rm{Irrigation\; Intensity}}}}=\frac{{\mbox{Irrigation Water}}({{{{\rm{m}}}}}^{3}\,{{{{\rm{ha}}}}}^{-1})}{{\mbox{Yield}}({\mbox{t}}{{{{\rm{ha}}}}}^{-1})}$$

(1-4)

where Irrigation Water represents the total volume of irrigation water applied per hectare (m³ ha⁻¹), and Yield is the grain output per hectare.

All three indicators are output-based resource consumption metrics, reflecting the agricultural system’s reliance on key inputs. Higher values indicate greater input consumption per unit of output and, consequently, higher resource dependency. Lower values suggest more efficient or conservative resource use.

Combined with the temporal evolution of cropland landscape indexes, these indexes provide a quantitative basis for assessing whether cropland consolidation contributes to improved resource use efficiency. Their integration helps to elucidate the role of landscape configuration in facilitating the transition toward more sustainable and intensive agricultural systems.

Evaluation of sustainable intensification of agriculture

By analyzing the number of regions in each combination, we further assessed the status of sustainable intensification of agriculture. Spatially, multiple pattern combinations may coexist in each region. All pattern combinations in each region were counted and scores of these combinations were summed up to qualify the degree of sustainable intensification of agriculture. We used equal interval classification method to divide the scores into six grades (i.e., A⁺, A, B, C, D, E, F). Grade A⁺ and A denotes that one region has achieved high degree of sustainable intensification of agriculture in terms of environmental protection and mechanical energy saving, demonstrating great resource use efficiency and effective environmental protection measures. It should be noted that grades D, E, and F indicate a state of unsustainable agricultural intensification in the region, with clear challenges in resource-use efficiency and environmental protection.

Potential for sustainable agricultural intensification

Cropland concentration and continuity contribute to the sustainable intensification of agriculture. In this study, we defined “farm size” as the area of cropland that has already achieved scale operation. In contrast, “cropland size” refers to the area of contiguous cropland patches. The difference between farm size and cropland size reflects the potential for the transformation of large-scale cropland to a scaled operation. The scale of farms in China has remained relatively stable over the past few decades due to the implementation of the Household Responsibility System in 1978⁸⁴, Therefore, we have opted to compare the farm size dataset provided by Lesiv et al.⁷⁵. with our calculated cropland dataset for 2020, in order to evaluate the potential for sustainable agricultural intensification in China in the future. This is also the most accurate global farm size data available to date. In Lesiv et al.’s study, they divided the farm size into five categories according to the area (namely: <0.6 ha, 0.6 ~ 2.6 ha, 2.6 ~ 16 ha, 16 ~ 100 ha, and >100 ha), and analyzed the spatial distribution of each category.

Future scenario projection

Based on the calibrated Markov–CA framework, we projected China’s cropland pattern to 2040 under a business-as-usual scenario. The simulation shows that provincial cropland configurations remain largely stable, with only slight changes in patch size, fragmentation, and aggregation. The detailed model settings, parameter sensitivity analysis, and provincial landscape indexes are provided in the Supplementary Materials (Supplementary Note 7).