Approximately three-quarters of the Earth's land surface has been altered by humans within the last millennium. Addressing global sustainability challenges like climate change, biodiversity loss, and food security hinges on understanding land use change, as it significantly impacts carbon sources and sinks, causes habitat loss, and underpins food production. The mitigation potential of land use activities is crucial for meeting climate targets under the Paris Agreement, making it a central topic in international policy debates. Accurately quantifying and understanding global land use change and its spatiotemporal dynamics is therefore critical. However, despite its societal relevance, comprehending global land use/cover (LUC) change across space and time is hampered by a lack of comprehensive data and significant uncertainties in existing LUC reconstructions. Even with the advancements in satellite technology, big data, and open access initiatives, LUC data remain fragmented, vary in scale and detail, and often lack consistent time series. Satellite remote sensing excels in high spatial resolution but offers short temporal coverage of land cover (biophysical properties). Inventories and statistics focus on land use (human activities on land) and provide long time spans but lack spatial detail due to their reliance on administrative units. Each data source independently lacks a critical component (space, time, or theme), hindering the full capture of land use dynamics. Existing global, long-term land use reconstructions often rely on limited observational data streams and make assumptions about factors like cropland allocation or wood harvests. They also suffer from coarse spatial resolutions and limited land use categories. While recent studies have improved spatial and temporal resolution, they often focus solely on land cover and not land use, or lack the capacity to fully account for gross change (all land transitions between LUC categories). This limitation is crucial when assessing the environmental and climatic impacts of LUC change. To address these limitations, this study combined multiple, high-resolution remote sensing data with long-term statistical data streams to assess annual LUC changes from 1960 to 2019 at a 1 km spatial resolution.

The authors review existing global, long-term land use reconstructions, highlighting their limitations. These include reliance on a few observational data streams, assumptions concerning allocation of cropland and wood harvests, coarse spatial resolution (up to 0.25 degrees), and limited land use categories. While acknowledging progress made by datasets like GLASS-GLC in assessing long-term land cover change at high resolution (5km) and temporal coverage (1982-2015), they emphasize that GLASS-GLC focuses on land cover, not land use, and relies on a single satellite sensor. Most importantly, the review points out that none of the existing datasets fully account for gross changes—all land transitions between LUC categories—which are critical for quantifying the climatic and environmental impacts. The authors cite several specific examples of previous studies (e.g., HYDE3.2, LUH2, SAGE cropland) to illustrate these limitations and the gaps in existing data. The lack of comprehensive data on gross changes and the varying scales and inconsistencies across different datasets necessitate the need for a more robust and detailed approach to studying global land use change.

To comprehensively analyze spatiotemporal dynamics of global land use change, the researchers developed the Historic Land Dynamics Assessment + (HILDA+) model. HILDA+ integrates multiple high-resolution remote sensing data (detailed in Supplementary Table 1) with long-term statistical data streams (FAO land use and population data). This innovative model assesses annual changes in LUC from 1960 to 2019 at a 1 km spatial resolution. It harmonizes spatially explicit LUC information with national-scale land use inventories and allocates these changes to the global land surface using open datasets. The approach fully incorporates data-derived, annual gross changes between six LUC categories: urban, cropland, pasture/rangeland, forest, unmanaged grass/shrubland, and sparse/no vegetation (Supplementary Table 2). This detailed approach allows for quantification of land use change's spatial extent and tracking of annual dynamics. The methodology involved several key steps: **Pre-processing of remote sensing-based LUC data:** The HILDA+ reconstruction utilized multiple openly available global, continental, regional, and national LUC datasets (Supplementary Table 1). **Harmonization of LUC maps:** A common generalized classification scheme (based on FAO land use and LCCS land cover schemes) was implemented to harmonize the different remote sensing products. Maps were reclassified, converted into binary masks, reprojected, and resampled to a 1x1 km grid resolution. For years lacking data, a linear extrapolation method was employed using long time series remote sensing products. **Probability maps for LUC categories:** Maps of average area fractions were generated for each land cover category, creating probability maps for the final six LUC categories (Supplementary Table 5). Additional data like Gridded Livestock World v3 (GLW) were used to refine the probability layers for pasture/rangelands. **Base map calibration:** The Copernicus LC100 Global Land Cover map for 2015 was used as the base map, calibrated using FAO national land use statistics for forest, cropland, and pasture area. **Preparing datasets for national LUC change matrices:** FAO land use area and population statistics were compiled per country and year, with data gaps filled by linear temporal interpolation and extrapolation. Country-specific gross change ratios were derived from transition matrices using remote sensing data. **Change calculation:** Net changes in cropland, pasture/rangeland, and forest were calculated from FAO data, applied to the base map. Urban area change was estimated using population data. The remaining land was divided proportionally between unmanaged grass/shrubland and sparse/no vegetation. **Change allocation:** A new transition matrix, incorporating all gross changes, was created for each time step, country, and land transition. Change magnitude was distributed using probability maps. This involved three rounds of allocation, prioritizing grid cells with higher area fractions of the relevant LUC category. **Change analysis:** Annual maps of LUC states and transitions allowed for the analysis of spatial extent, patterns, rates, and dynamics of land use change. Changes were categorized into gain (single), loss (single), and multiple change events. **Uncertainty assessment:** Annual uncertainty layers were generated based on input datasets, considering the number of datasets, maximum deviation in area fraction, and mean class area fraction. These layers reflect data quality and agreement.

HILDA+ estimates that 17% of the Earth's land surface changed at least once between 1960 and 2019. Considering all change events (including multiple changes within the same area), the total land change extent reached 43 million km², almost a third of the global land surface. This represents an average annual land area change equivalent to twice the size of Germany. The study reveals a net loss of global forest area (0.8 million km²) and expansion of global agriculture (cropland: 1.0 million km², pasture/rangeland: 0.9 million km²). However, these global trends mask significant regional variations. Forest areas in the Global North (including China) increased, while those in the Global South decreased considerably. The opposite pattern is observed for cropland—a decrease in the Global North and an increase in the Global South. Changes in pasture/rangeland are less distinctly North-South divided, with both China and Brazil significantly contributing to global expansion. The analysis distinguishes between single change events (e.g., deforestation) and multiple change events (e.g., crop-grass rotation). Single change events (38% of all transitions) are prevalent in developing countries of the Global South, often representing agricultural expansion (e.g., pastureland expansion in China, tropical deforestation in the Amazon). Multiple change events (62% of transitions) dominate in developed countries of the Global North and rapidly growing economies, frequently related to agricultural intensification (e.g., Europe, the US, Nigeria, India). Temporal analysis reveals two distinct phases of land use change: (1) an acceleration phase (1960–2004) with increasing rates, and (2) a deceleration phase (2005–2019) with decreasing rates. The acceleration phase coincides with a shift from agro-technological intensification to production for globalized markets and increasing trade, particularly impacting the Global South. The deceleration phase is hypothesized to be linked to the global economic and food crisis of 2007–2009, reduced demand for commodities, and decreased large-scale land acquisitions. Other factors, like climate change and extreme events, also played a role in the deceleration phase. Comparison with existing land use reconstructions (LUH2, HYDE3.2, SAGE cropland) demonstrates that the area affected by global land use change in the HILDA+ study is almost four times greater. This difference stems from HILDA+'s inclusion of gross changes derived from Earth observation data. Comparison with other high-resolution remote sensing datasets (Hansen GFC, ESA CCI, MODIS) shows similar magnitudes of change, although specific annual rates vary due to differences in land cover classifications and semantics.

The findings highlight that the extent of global land use change over the past six decades is substantially higher than previously recognized. The identification of diverging regional patterns underscores the complexity of land use change dynamics, emphasizing the importance of considering geographical context. The study demonstrates a strong link between global trade and agricultural production as a primary driver of land use change, particularly deforestation for commodity crops in the Global South. The observed shift from accelerating to decelerating land use change after the 2007-2009 economic crisis suggests a significant influence of market mechanisms and global economic trends. While acknowledging the role of other factors like climate change, the study emphasizes the crucial impact of global trade and economic fluctuations on land use patterns. The improved estimates of land use change provided by HILDA+ have significant implications for assessing climate change, biodiversity loss, and food security, particularly in carbon budget estimations and forest management strategies. The increased precision in quantifying land use change offers enhanced opportunities for analyzing temporal trends, identifying potential drivers and impacts, and informing policy decisions related to global sustainability goals.

This study, using the HILDA+ model, presents a significantly refined estimate of global land use change, indicating a scale four times greater than previously assumed. The integration of multiple data sources provides a more comprehensive and spatially detailed understanding of land use dynamics. The identified diverging patterns between the Global North and South, and the influence of global trade on land use change, highlight the complexities of this global issue. This research offers crucial improvements to the assessment of climate change, biodiversity loss, and food security. Future research should focus on refining the model's uncertainty estimates and incorporating additional data sources to further enhance its accuracy and detail. The continuous refinement and integration of HILDA+ data with other models and assessments will be critical for improving land-use policies and achieving global sustainability goals.

While HILDA+ represents a significant advance, certain limitations exist. The accuracy of the model depends on the quality and availability of input data, which may vary across regions and time periods. Data ambiguities and inconsistencies across datasets, especially in heterogeneous landscapes, can influence the results. The model's reliance on existing datasets propagates existing uncertainties in these datasets. While efforts were made to harmonize and calibrate data, variations in classification schemes and land cover definitions can still affect the accuracy of land use estimations. Although HILDA+ addresses many of the limitations of previous models by utilizing multiple data streams, the interpretation of results should still be mindful of potential uncertainties and regional variations in data quality and resolution.