Global maps of cropland extent and change show accelerated cropland expansion in the twenty-first century

Global population growth and rising living standards necessitate expansion and intensification of agricultural land use to meet increasing demands for food, biofuels, and other commodities. This expansion and intensification, however, pose significant threats to ecosystem functioning and biodiversity through habitat loss and fragmentation. The UN's 2030 Sustainable Development Goals (SDGs) emphasize the need to balance increased agricultural production with the preservation of ecosystem services. Achieving these goals requires consistent, independent, and timely data on agricultural extent and productivity to inform national policies and international cooperation. Spatiotemporally consistent satellite observations offer the most accurate and cost-effective method for global agricultural land-use mapping and monitoring. While satellite data has been used for national and global agriculture mapping, a globally consistent, multidecadal, high-resolution (30m per pixel) cropland time-series dataset has been lacking until now. This study aims to address this gap by presenting a global cropland extent and change dataset to monitor progress towards the SDGs. Cropland is defined as land used for annual and perennial herbaceous crops for human consumption, forage, and biofuel, excluding perennial woody crops, permanent pastures, and shifting cultivation (fallow limited to 4 years). This definition aligns closely with the UN Food and Agriculture Organization's (FAO) arable land category.

Existing research highlights the interconnectedness of food security, agricultural intensification, and biodiversity loss. Studies like Godfray et al. (2010) and Tilman et al. (2011) emphasize the challenges of feeding a growing population while maintaining sustainable agricultural practices. Foley et al. (2005) and Gibbs et al. (2010) have documented the significant impacts of land use change, particularly deforestation, on global ecosystems. The UN's SDGs underscore the urgency of balancing these competing demands. Previous efforts in global agricultural mapping, utilizing satellite data, have shown promise (Ramankutty et al., 2008; Hu et al., 2020; Boryan et al., 2011; Pittman et al., 2010; Buchhorn et al., 2020), but lacked the consistent, high-resolution temporal coverage needed for comprehensive analysis. This study builds upon these previous efforts by providing a more detailed and spatially explicit dataset.

This study utilizes a consistently processed 30m spatial resolution Landsat satellite data archive from 2000 to 2019 to create a global cropland extent and change dataset. The Landsat time-series data were transformed into multitemporal metrics characterizing land surface phenology using a machine-learning classification to map global cropland extent. The classification models were locally calibrated using extensive training data collected by visual interpretation of high-spatial-resolution satellite images. A probability sample, stratified based on Landsat-based global cropland maps, was used to estimate cropland area and its associated uncertainty, analyzing land-use conversion pathways. Sample reference data were collected through visual interpretation of Landsat time-series data and higher-resolution satellite images. Cropland maps were integrated with Moderate Resolution Imaging Spectroradiometer (MODIS)-derived annual net primary production (NPP) as a proxy for crop productivity. The analysis was performed in 4-year epochs (2000–2003, 2004–2007, 2008–2011, 2012–2015, and 2016–2019), generating one cropland map per epoch. The methodology involved three stages for global cropland mapping: 1) individual cropland classifications for a set of Landsat ARD 1° × 1° tiles for the 2016–2019 interval; 2) using these classified tiles to train regional cropland mapping models and applying them to preceding 4-year intervals; and 3) using the preliminary global cropland maps as training data to generate temporally consistent global cropland data. Bagged decision tree ensembles were used as the supervised classification algorithm. A sample analysis was conducted to estimate cropland area and uncertainty, and assess map accuracy for 2003 and 2019. The sample-based area estimation utilized a stratified random sampling approach. Map accuracy metrics included overall accuracy, user's accuracy, and producer's accuracy. Cropland NPP was evaluated using the MODIS-based annual year-end gap-filled NPP product (MOD17A3HGF). For national-level analysis, publicly available GIS country boundaries and population data were used to calculate per-capita cropland area and NPP.

The study estimated the 2019 global cropland area to be 1,244.2 ± 62.7 Mha. From 2003 to 2019, global cropland area increased by 101.9 ± 45.1 Mha (9% of the 2003 area), with the largest expansion in Africa (34%) and South America (49%). Global cropland expansion accelerated, nearly doubling the annual expansion rate from 5.1 MHa per year to 9.0 MHa per year. Half (49%) of new cropland replaced natural vegetation. Global per-capita cropland area decreased by 10% due to population growth, but per-capita annual cropland NPP increased by 3.5% due to intensification. Comparison with FAO arable land data showed good agreement, though the study's estimates were smaller, likely due to definitional differences. Analysis of land-use conversions revealed that half of new cropland replaced natural vegetation, primarily in Africa, South-east Asia, and South America. The other half resulted from pasture conversion and recultivation of abandoned land. Cropland loss was primarily due to abandonment/pasture conversion (52%), construction/infrastructure (16%), and conversion to other land uses (13%). National-level analysis showed the USA, India, and China had the largest cropland areas in 2019. Brazil and India experienced the largest net increases, while Russia and Cuba showed the largest reductions. The study found a 25% increase in global cropland NPP from 2003 to 2019, primarily due to area expansion and increased productivity per unit area. The highest NPP increases were observed in South America and Africa. Regional accuracy varied, with higher accuracy in regions dominated by large-scale industrial farming.

The findings highlight the significant and accelerating expansion of global cropland, particularly in Africa and South America, driven by population growth and increasing food demand. The replacement of natural vegetation with cropland underscores the trade-off between food production and ecosystem conservation. The increased cropland NPP reflects both expansion and intensification of agricultural practices. The study's high-resolution data provide a valuable tool for monitoring land-use change at various scales, informing sustainable development policies. The observed regional differences in cropland dynamics reflect diverse factors such as agricultural practices, land tenure systems, and national policies. The comparison with FAO data highlights the importance of consistent definitions for accurate monitoring and reporting. The study’s limitations in capturing heterogeneous landscapes and the potential underestimation of NPP in some areas should be considered.

This study provides a high-resolution, globally consistent time-series of cropland extent and change from 2000 to 2019. The findings reveal an accelerating rate of cropland expansion, particularly in Africa, with substantial conversion of natural vegetation. While per capita cropland area has decreased, per capita NPP has increased due to intensification. This dataset is crucial for monitoring progress towards SDGs and informing sustainable land management strategies. Future research could focus on improving the accuracy of NPP estimation at finer spatial scales and integrating this data with socioeconomic factors to better understand the drivers of land-use change.

The study's accuracy is affected by spatial resolution limitations, especially in heterogeneous landscapes, leading to potential underestimation of cropland area in some regions. The use of MODIS NPP data, which has a coarser resolution than the Landsat data, may also limit the accuracy of the NPP estimates, particularly in heterogeneous areas. The definition of cropland used in the study may differ from other classifications, potentially affecting comparability with other datasets. Finally, the analysis relies on satellite imagery, which may be affected by cloud cover and other factors, potentially introducing some uncertainty in the results.