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

The study addresses the need for consistent, high-resolution, global information on cropland extent and change to support sustainable food production while safeguarding ecosystems. Global population growth and rising living standards are driving the expansion and intensification of agricultural land use to meet increasing demand for food, biofuel and other commodities, but expansion threatens ecosystem functioning and biodiversity through habitat loss and fragmentation. Achieving the UN 2030 Sustainable Development Goals requires nationally and internationally coordinated policies grounded in accurate, timely, and independent data on agricultural extent and productivity. While satellite observations enable accurate and cost-effective global land-use mapping, a globally consistent, multidecadal cropland time-series at locally relevant spatial resolution (30 m) has been lacking. The authors present a new global cropland extent and change dataset (2000–2019) designed to fill this gap and to monitor progress towards SDGs, defining cropland as land used for annual and perennial herbaceous crops for human consumption, forage (including hay), and biofuel, excluding perennial woody crops, permanent pastures, and shifting cultivation, with a maximum fallow length of 4 years.

Prior work shows satellite data can support national and global agriculture mapping and monitoring, and that global cropland expansion and intensification impact ecosystem services and biodiversity. Existing products (for example, FAO arable land statistics, MODIS-based cropland and Copernicus land cover) provide valuable insights but lack a globally consistent, multidecadal 30 m cropland time-series at local relevance. Studies have documented cropland expansion impacts (for example, forest conversion) and demonstrated the utility of MODIS and Landsat time-series for cropland mapping and productivity assessment. However, limitations include coarse spatial resolution of productivity products and definitional inconsistencies across reported statistics. This study builds on these by providing a harmonized Landsat-based global cropland map time-series and probability-based area estimates to reconcile and benchmark against FAO statistics.

Scope and temporal design: Global cropland mapping was performed over a domain defined by the USGS GFSAD 30 m masks using a 1° × 1° tiling scheme, excluding small islands without geometrically corrected Landsat data. Five 4-year epochs were mapped: 2000–2003, 2004–2007, 2008–2011, 2012–2015 and 2016–2019. A pixel is mapped as cropland for an epoch if a growing crop is detected in any year within that 4-year interval, implementing a maximum fallow criterion of 4 years. Input data and preprocessing: The study used globally normalized Landsat Analysis Ready Data (ARD) 16-day surface reflectance composites (1997–2019), normalized to MODIS surface reflectance, with haze/cloud/shadow exclusions and linear interpolation for missing 16-day intervals. For each epoch, annualized gap-free 16-day time-series were constructed from 4 years of observations by selecting the highest NIR reflectance per interval to prioritize vegetated observations. Time-series were transformed into multitemporal metrics capturing land-surface phenology (ranks, inter-rank averages, amplitudes, reflectance and vegetation index characteristics for phenological stages) and augmented with elevation data. Classification pipeline: A three-stage machine-learning workflow using bagged decision tree ensembles at the Landsat pixel scale was implemented. Stage 1: per-tile supervised classifications for 924 representative 1° × 1° tiles for 2016–2019 using manually interpreted training data (cropland presence/absence) from Landsat metrics and high-resolution imagery. Stage 2: regional models trained from the 924 tile results were iteratively refined and applied to all epochs to produce a preliminary time-series. Stage 3: temporally consistent local models were created by using preliminary maps as training within a 3° radius for each of 13,451 tiles across all epochs, prioritizing stable cropland/non-cropland. A single ensemble per tile was calibrated with training from all intervals, applied to each epoch, and post-processed to remove isolated class flickers, eliminate patches <0.5 ha, and mask artefacts (e.g., overestimation in wetlands/flooded grasslands). A per-pixel cropland probability threshold of 0.5 produced binary maps. Sample-based estimation and accuracy assessment: Probability-based stratified random sampling (five strata per region: stable cropland, cropland gain, cropland loss, possible omission, other lands) was conducted for 2003 and 2019 in seven global regions and globally (100 samples per stratum; 3,500 total). Two experts independently interpreted each sample using Landsat ARD time-series, multitemporal composites, and high-resolution imagery; disagreements were resolved by consensus. Sample-based estimators quantified total cropland area, stable cropland, gross loss, gross gain, and net change with standard errors and 95% confidence intervals. Ratio estimators quantified proportions of land-use trajectories within gain and loss. Map accuracies (overall, user’s, producer’s) and uncertainties were computed using stratified estimators. Productivity analysis: MODIS MOD17A3HGF annual gap-filled NPP (500 m) was resampled to the Landsat grid and intersected with cropland maps to compute total and per-unit-area cropland NPP for each epoch (3-year average for 2001–2003; 4-year averages thereafter). Per-capita metrics used UN World Population Prospects 2019 to relate population (2003, 2019) to cropland area and cropland NPP at global, regional, and national scales. National analyses used GADM country boundaries. Data and codes are publicly available via GLAD UMD and LP DAAC portals.

• Global cropland area in 2019 was 1,244.2 ± 62.7 Mha (95% CI). Regional shares of 2019 cropland: Eurasia (Europe and North Asia plus South-west Asia) 55%, Africa 17%, North and Central America 16%, South America 9%, Australia and New Zealand 3.
• From 2003 to 2019, global cropland area increased by 101.9 ± 45.1 Mha (9% of 2003 area). Africa had the largest absolute gain (53.2 ± 39.4 Mha; 34%), and South America had the largest relative gain (37.1 ± 8.7 Mha; 49%). Australia and New Zealand and South-west Asia showed moderate expansion; North/Central America, Europe/North Asia, and South-east Asia had small net changes with substantial offsetting gross gains and losses.
• Global cropland expansion accelerated: the map-based annual expansion rate nearly doubled over two decades (world total: 5.1 Mha yr⁻¹ in 2004–2007; 6.3 in 2008–2011; 10.9 in 2012–2015; 9.0 in 2016–2019). Africa’s expansion rate more than doubled by 2016–2019 (from 1.7 to 3.9 Mha yr⁻¹), while South America’s rate declined by 2016–2019 (from 2.7 to 1.5 Mha yr⁻¹).
• New cropland since 2003 represents 217.5 ± 37.7 Mha (17%) of the 2019 cropland area; the highest proportions of new cropland were in South America (39%) and Africa (34%).
• Land-use conversions for cropland gain: 49% of gross cropland gain replaced natural woody/herbaceous vegetation or tree plantations (global; Table 2), including 5% via dryland irrigation (notably in South-west and South-east Asia and North America). Africa (excluding dryland irrigation) had 79% of gains from natural vegetation conversion; South-east Asia 61%; South America 39%. The remaining 51% of gains came from conversion of pastures and recultivation of abandoned agricultural lands (e.g., 97% in Europe/North Asia; 91% in Australia/New Zealand; 75% in North and Central America; 61% in South America).
• Cropland losses affected 115.5 ± 24.1 Mha (10% of 2003 cropland). Major loss trajectories: abandonment or conversion to pastures (52%); construction/infrastructure/mining (16%; up to 35% of losses in South-east Asia due to urbanization); conversion to other intensive agriculture (permanent woody crops, aquaculture) (13%; 28% in South-east Asia); flooding/water bodies (3%); restoration to natural vegetation or tree plantations (16%).
• Population and per-capita metrics: Global population rose 21% (6.4 to 7.7 billion) from 2003 to 2019. Per-capita cropland area fell 10% (0.18 to 0.16 ha person⁻¹); only South America increased per-capita cropland. Despite area decrease, global per-capita cropland NPP increased by 3.5% due to intensification.
• Productivity: Cropland NPP increased 25% globally (4.43 to 5.53 Pg C yr⁻¹ from 2001–2003 to 2016–2019). South America had the largest relative NPP increase (0.38 Pg C yr⁻¹; 88%), followed by Africa (0.29 Pg C yr⁻¹; 50%). Mean NPP per unit area in stable cropland rose ~10% (402 to 442 g C m⁻² yr⁻¹), with South America’s stable cropland NPP rising 25% (528 to 730 g C m⁻² yr⁻¹), accounting for 34% of the total global cropland NPP increase.
• Validation and comparisons: Sample-based cropland area agrees well with FAO arable land (R² = 0.94 in 2003; 0.98 in 2019), though FAO totals are larger (by 16% in 2003; 11% in 2019), reflecting definitional differences. National 2019 cropland areas correlate strongly with FAO 2018 arable land (R² = 0.97) and Copernicus 100 m cropland fraction (R² = 0.96), with noted over/underestimation in specific countries due to land-use and mapping differences. The USA had the largest cropland area in 2019, followed by India and China; largest net increases occurred in Brazil (+23.1 Mha; +77%) and India (+15.5 Mha; +13%); largest reductions in Russia (−5.7 Mha; −6%) and Cuba (−0.5 Mha; −28%).

Findings demonstrate that cropland expansion has accelerated globally in the early 21st century, particularly in Africa, while decelerating in South America in recent years. Nearly half of new cropland replaces natural vegetation, underscoring trade-offs with biodiversity conservation and SDG 15 targets to halt deforestation and habitat degradation. The per-capita decline in cropland area, offset by rising per-unit-area productivity (NPP), indicates intensification as a key driver supporting food supply amid population growth. Spatial patterns of cropland distribution and dynamics transcend political boundaries, aligning with agro-ecological potential, population pressure, and land-use history (e.g., Great Plains, Pampas, Pontic steppe, Indo-Gangetic Plain). Regional narratives include synchronized expansion across Brazil–Paraguay–Bolivia–Uruguay and Sahel/Central Africa, dryland irrigation expansion and urban/agro-industrial transitions in parts of Asia, and post-Soviet abandonment with recent recultivation in southern Russia. National-scale analyses reveal variable capacity to expand cropland relative to population growth; some African nations saw per-capita stability due to large area gains, while others experienced substantial per-capita declines, signaling potential food insecurity risks. The strong agreement with FAO and Copernicus benchmarks supports the robustness of estimates while highlighting definitional and mapping differences. The produced 30 m time-series enables stratified sampling, local-to-global monitoring, and integration with other environmental change datasets (forest loss, water dynamics) to support SDG tracking and climate-related agricultural emissions estimation.

The study delivers the first spatiotemporally consistent, 30 m global cropland extent and change maps for five epochs from 2000 to 2019, accompanied by probability-based area estimates and integration with MODIS NPP for productivity assessment. Results show a 9% increase in global cropland area, accelerated expansion rates (especially in Africa), substantial conversion from natural vegetation, and a 25% rise in cropland NPP with modest per-capita NPP gains despite per-capita area decline. The openly available maps and reference sample data provide critical, locally relevant inputs for monitoring natural land appropriation, assessing progress towards SDGs, supporting agricultural decision-making, and improving higher-resolution NPP and yield modeling. Future work should refine change detection accuracies, address heterogeneous landscapes, harmonize definitions with statistical reporting, integrate higher-resolution productivity datasets and crop-type information, and link cropland dynamics more directly to yields, greenhouse gas emissions, and biodiversity outcomes.

• Mapping accuracy varies by region: highest in North and South America with large-scale farming; underestimation of cropland in heterogeneous landscapes of Europe, Asia, and Africa due to 30 m spatial resolution; overestimation in Australia/New Zealand from inclusion of intensively managed permanent pastures.
• Change detection is more challenging than static mapping, with generally lower accuracies across regions.
• Definitional differences with FAO statistics (e.g., inclusion of unused arable land or other agricultural uses in FAO) lead to discrepancies; this product maps actively cultivated cropland with a maximum 4-year fallow.
• Productivity assessment uses MODIS NPP (500 m), which tends to underestimate NPP, especially for irrigated crops, and its coarse resolution relative to Landsat (30 m) can hinder analyses in heterogeneous areas.
• Sample-based accuracy and area estimation were conducted only for 2003 and 2019; intermediate epochs were assumed to have similar accuracies due to consistent processing.
• Some regions required manual masking to remove artefacts (e.g., wetlands, flooded grasslands); small islands were excluded due to lack of suitable Landsat geometry.
• National differences in land-use reporting and undocumented abandonment (e.g., Russia) affect comparisons with external datasets.