Global land use changes are four times greater than previously estimated

The paper addresses how global land use/cover (LUC) has changed across space and through time, a key issue for climate mitigation, biodiversity conservation, and food security. Despite the importance, current understanding is limited by fragmented datasets, inconsistent time series, varying scales, and a disconnect between land cover products (high spatial resolution, short temporal coverage) and land use statistics (long temporal coverage, coarse spatial detail). Existing reconstructions (e.g., HYDE3.2, LUH2, SAGE cropland) have coarse resolution, limited categories, and often rely on assumptions, and they typically omit gross changes (all transitions within a period). The study’s purpose is to produce a comprehensive, spatially explicit, long-term reconstruction of global LUC change that fully accounts for gross transitions, enabling improved quantification of environmental and climatic impacts and insights into drivers and temporal dynamics.

The authors review limitations of established datasets and approaches: HYDE3.2 and LUH2 rely on assumptions for cropland allocation and wood harvests and have coarse spatial resolutions (up to 0.25°) and limited land use categories. SAGE provides cropland estimates but lacks broader LUC categorization. GLASS-GLC advanced long-term land cover at 5 km resolution (1982–2015) but is land cover only and relies on a single sensor (AVHRR). Critically, most prior reconstructions fail to capture gross LUC changes, underestimating the extent of transitions that matter for climate and environmental assessments. The review also cites regional studies evidencing divergent trends (e.g., reforestation and agricultural abandonment in parts of the Global North; commodity-driven deforestation in the Global South), reinforcing the need for a harmonized, global, gross-change dataset.

The study develops HILDA+ (Historic Land Dynamics Assessment+) to reconstruct annual global LUC changes (1960–2019) at 1 km resolution across six categories: urban, cropland, pasture/rangeland, forest, unmanaged grass/shrubland, and sparse/no vegetation. Key steps: - Data sources: multiple remote sensing products (global, regional, national), FAO land use (arable and permanent cropland, permanent meadows/pastures, forest) and population statistics, and supporting datasets (e.g., GLW ruminant density), all harmonized to a common scheme. - Harmonization: reclassify heterogeneous land cover products into six generalized classes (aligned with FAO and LCCS definitions), convert to binary masks, reproject to Eckert IV at 1×1 km, and derive area fraction maps. For years without observations, back-cast time series (e.g., ESA CCI, MODIS, GLAD UMD VCF) via linear extrapolation of early trends. - Probability maps: average area fractions across available datasets per year to create probability layers for each class. Managed vs unmanaged grass/shrubland separation uses ruminant density (GLW v3, 2010) combined with grass/shrub fractions to indicate pasture/rangeland probabilities; unmanaged grass/shrubland derived from grassland+shrubland averages. - Base map (2015): Copernicus LC100 2015, reclassified and resampled, calibrated to FAO national areas (forest, cropland, pasture/rangeland) using derived area fractions. - National transition matrices: compile FAO land use and population by country/year (harmonized to 2015 borders, gap-filled via linear interpolation/extrapolation; for Europe, supplemented with HILDA predecessor where FAO lacks records). Derive country-level gross change ratios from mean transition matrices computed from temporally consistent land cover maps (ESA CCI and regional products such as CORINE, MOEF Indonesia, AAFC Canada, NLCD USA, Australia DLCD). - Change calculation: compute net changes for cropland, pasture/rangeland, forest from FAO relative changes applied to base map areas; urban net change proxied by population development; residual land allocated between unmanaged grass/shrubland and sparse/no vegetation proportionally to base map ratios. - Iterative change allocation: for each country, year, and transition, allocate gross change magnitudes across the grid using probability maps via three rounds (round 1: assign where target class has highest area fraction >0.1; round 2: if needed, area fraction >0.4; round 3: no allocation if still unmet). The process runs backward (2015–1960) and forward (2015–2019), producing annual global LUC maps and transition layers. - Change analysis: classify transitions into gain, loss, and multiple-change events per category; compute frequencies and spatial extents; aggregate globally and regionally. - Uncertainty assessment: per-pixel quality metrics from number of datasets, maximum deviation and mean class area fraction per year; global quality flags mapped to indicate areas of higher/lower agreement. - Implementation: Python 3.7; data distributed via PANGAEA and interactive map viewer; code available upon request.

- Extent of change: 17% of Earth’s land surface changed at least once between 1960–2019. Summing all gross transitions yields 43 million km² of change events, almost one-third (≈32%) of global land area. On average, about 720,000 km² (≈twice Germany) changed per year since 1960. - Category balances: global net forest loss of 0.8 million km²; cropland expanded by 1.0 million km²; pasture/rangeland expanded by 0.9 million km². - Regional divergence: forest area increased in the Global North (including China) but decreased strongly in the Global South. Cropland declined in the North and increased in the South. Pasture expansion notable in China and Brazil. - Single vs multiple transitions: 38% of transitions are single-change events, predominating in the Global South; about 48% of these single events are agricultural expansion (e.g., Amazon deforestation, Chinese pasture expansion). Multiple-change events constitute 62% and dominate in the Global North and fast-growing economies (e.g., EU, USA, Australia, Nigeria, India). Of multiple-change events, 86% are agricultural transitions; 11% are cropland–pasture/rangeland switches (indicative of rotations/mixed systems); 75% occur between managed and unmanaged lands (e.g., cropland abandonment, shrub encroachment, agroforestry dynamics). - Temporal dynamics: two phases—acceleration of land use change from 1960 to 2004/2005, followed by deceleration from 2005 to 2019. Acceleration is strongest in the Global South (South America, Africa, Southeast Asia) alongside rising commodity crop production and exports. Deceleration correlates with the 2007–2009 global economic/food crisis: pre-crisis surging demand (food, feed, biofuels) and high oil prices (peak $145.31 in 2008) drove expansion; post-crisis demand dropped, large-scale land acquisitions declined, and agricultural expansion slowed, especially in Argentina, Brazil, Ghana, Indonesia, and Ethiopia. Climate extremes (e.g., 2010/11 drought in Ethiopia) further influenced declines in change rates. - Category-specific dynamics: forest shows steady net annual decrease with acceleration in the 1990s; cropland and pasture/rangeland exhibit larger interannual variability (≈4× forests), reflecting rapid socio-economic responses (policy shifts, trade disruptions, conservation incentives, extreme events). Pasture/rangeland shows a downward trend linked to livestock technology advances, whereas cropland experienced renewed expansion waves since 2000. - Comparison with other datasets: HILDA+ indicates global LUC change is ~3.7× greater than long-term reconstructions previously suggested. Mean annual gross change rate comparisons (10³ km² a⁻¹): LUH2 302±125 vs HILDA+ 721±88 (1960–2015; HILDA+ 2.4× LUH2); HYDE3.2 cropland 187±82 vs 246±41; HYDE3.2 pasture 57±25 vs 420±71 (HILDA+ 4.4× HYDE overall); SAGE cropland 203±74 vs 253±37; ESA CCI (all with combined grassland) 249±165 vs 578±40 (1992–2015); MODIS (all with combined grassland) 1123±44 vs 574±43 (2001–2015); Hansen GFC forest gain 265±27 vs 270±21 (2000–2012). Remote sensing change rates are on the same order as HILDA+ (on average ~1.1×); deviations vary by product (MODIS +90%, ESA CCI −60%, Hansen ~0% relative to HILDA+). - Uncertainty: Highest dataset agreement for forests and sparse/no vegetation; larger deviations for cropland and pasture/rangeland in heterogeneous landscapes (e.g., Sub-Saharan savannahs, Australian rangelands, Central Asian steppes, Siberian taiga–tundra transition).

The findings address the central question of how global land use has changed in space and time by providing a harmonized, high-resolution, gross-change reconstruction. HILDA+ reveals that land use change is far more extensive than previously estimated and is characterized by strong regional contrasts: reforestation and agricultural abandonment in much of the Global North versus deforestation and agricultural expansion in the Global South. The temporal pattern of acceleration until the mid-2000s and subsequent deceleration aligns with shifts in globalized agricultural markets, trade teleconnections, biofuel policies, and the 2007–2009 economic/food crisis, indicating that market dynamics strongly modulate land change. Climate variability and extremes further modulate rates, particularly in drought-prone regions. The prevalence of multiple-change events in developed regions reflects intensification and complex management, underscoring the importance of capturing gross transitions to assess environmental impacts (e.g., carbon fluxes, habitat dynamics). By synthesizing multiple datasets and harmonizing them with national inventories, HILDA+ reduces individual data biases and provides consistent annual time series, enabling improved assessments of carbon budgets, biodiversity impacts, and food system dynamics and supporting policy targets (Paris Agreement, SDGs, post-2020 CBD agenda).

This study introduces HILDA+, a global, annual, 1 km resolution reconstruction of land use/cover change (1960–2019) that fully accounts for gross transitions across six categories. It shows that global land use change has affected nearly one-third of land over six decades—about four times greater than long-term reconstructions previously suggested—while revealing divergent regional trajectories and a pronounced acceleration–deceleration pattern linked to global trade and the 2007–2009 crisis. HILDA+ bridges gaps between spatially detailed but temporally limited remote sensing and temporally rich but spatially coarse inventories, providing uncertainty information and harmonized outputs. The dataset enables deeper analyses of drivers, impacts, and correlations of land change, improving estimates of carbon budgets, assessments of biodiversity loss, and evaluations of food security. Future research can leverage HILDA+ for global time-series analyses of land-use drivers, integrate additional management details (e.g., dynamic grazing intensity), refine uncertainty quantification in heterogeneous landscapes, and couple with Earth system and socio-economic models to evaluate mitigation/adaptation strategies and policy interventions.

- Data heterogeneity and uncertainties persist, especially for agricultural categories (cropland, pasture/rangeland) in heterogeneous landscapes (savannahs, rangelands, taiga–tundra ecotones), leading to larger dataset deviations. - Pasture/rangeland probability mapping uses static ruminant density data (GLW v3, 2010), not reflecting temporal changes in livestock distribution. - Back-casting of remote sensing products for years without observations relies on linear extrapolation of early trends, potentially introducing temporal biases. - Calibration and change estimation depend on national FAO statistics (including FRA forest data with 5-year reporting), which may contain reporting inconsistencies and limited temporal resolution. - Allocation rules (probability thresholds, multi-round assignment) and harmonization to six broad classes may mask finer management nuances and land cover distinctions. - Although multi-source harmonization attenuates single-dataset errors, HILDA+ is not free from artefacts; quality flags indicate variable confidence across regions and classes.