GLAD

The Global Land Cover and Land Use Change, 2000-2020

The GLAD Global Land Cover and Land Use Change dataset quantifies changes in forest extent and height, cropland, built-up lands, surface water, and perennial snow and ice extent from the year 2000 to 2020 at 30-m spatial resolution. The global dataset derived from the GLAD Landsat Analysis Ready Data. Each thematic product was independently derived using state-of-the-art, locally and regionally calibrated machine learning tools. Each thematic layer was validated independently using a statistical sampling. The global dataset is available online, with no charges for access and no restrictions on subsequent redistribution or use, as long as the proper citation is provided as specified by the Creative Commons Attribution License (CC BY). For all questions and comment contact Peter Potapov (potapov@umd.edu).

Dataset Reference

Potapov P., Hansen M.C., Pickens A., Hernandez-Serna A., Tyukavina A., Turubanova S., Zalles V., Li X., Khan A., Stolle F., Harris N., Song X.-P., Baggett A., Kommareddy I., Kommareddy A. (2022) The global 2000-2020 land cover and land use change dataset derived from the Landsat archive: first results. Frontiers in Remote Sensing [https://doi.org/10.3389/frsen.2022.856903] [Open PDF]

View Data in Google Earth Engine App

Combined land cover and land use change: https://glad.earthengine.app/view/glcluc-2000-2020

Data Download (GeoTIFF)

10x10 degree tiles of combined land cover for 2000, 2005, 2010, 2015 and 2020, and 2000-2020 land cover/use change can be downloaded from this page.

Forest Extent and Height Change, 2000-2020 

We defined forest as wildland, managed, and planted tree cover, including agroforestry and orchards. The forest height was mapped globally for woody vegetation of ≥ 3 m. We employed the global Landsat-based forest height model calibrated for the year 2019 using GEDI observations. For the boreal forests north of 52°N, where GEDI data are absent, we used a set of regional models calibrated with available GEDI data and manually collected training. The same model was applied to estimate forest height for the years 2000 and 2020. To reduce errors and noise in the model outputs we implemented extensive filtering of the output products. 

A set of global products were derived from the year 2000 and 2020 forest height maps:

Net forest height increase and decrease. We calculated the net forest height increase only if a pixel (i) demonstrated net forest extent gain, or (ii) has net forest height increase from the year 2000 to 2020 by ≥ 100%. Similarly, the net forest height loss was calculated only for forest loss pixels or in case the net height reduction was by ≥ 50% of the year 2000 value. Areas with small differences between the years 2000 and 2020 forest height data that have no indication of forest disturbance are considered stable forests in this prototype product.

Forest extent and change. We produced the year 2000 and 2020 forest extent maps by attributing pixels with ≥ 5 m forest height as the “forest” land cover class, to ensure consistency with the FAO FRA forest definition. The forest extent change (net forest extent loss and gain) was derived directly from the year 2000 and 2020 map comparison. Our forest definition differs from the one used by the FAO by the inclusion of trees outside forests (agroforestry, orchards, parks) and the exclusion of temporally unstocked forest areas.

Forest disturbance. Forests that have tree height ≥ 5 m in both years 2000 and 2020 and experienced significant or stand-replacement disturbance dosing the tile interval were assigned to “disturbed forest” class. The forest disturbance data for the 2001-2020 interval were combined from two datasets, the Global Forest Loss (GFL) data V1.8 and the new provisional annual global forest loss data.

Forest Height 2019 Dataset Reference

P. Potapov, X. Li, A. Hernandez-Serna, A. Tyukavina, M.C. Hansen, A. Kommareddy, A. Pickens, S. Turubanova, H. Tang, C.E. Silva, J. Armston, R. Dubayah, J. B. Blair, M. Hofton (2020) Mapping and monitoring global forest canopy height through integration of GEDI and Landsat data. Remote Sensing of Environment, 112165. https://doi.org/10.1016/j.rse.2020.112165 

Cropland Extent and Change, 2000-2003 and 2016-2019

Cropland was defined as land used to produce annual and perennial herbaceous crops for human consumption, forage, and biofuel (Potapov et al., 2021b). Our definition excludes tree crops, permanent pastures, and shifting cultivation. The cropland mapping was performed for two four-year epochs, 2000-2003 and 2016-2019. For each epoch, we attributed a GLAD ARD pixel as cropland if a growing crop was detected during any year within the interval. This mapping approach allowed us to implement the maximum fallow length rule. A global set of locally calibrated cropland mapping models that extrapolated manually delineated training areas in space and time was used to map cropland extent. The final products were filtered by removing artifacts and cropland patches smaller than 0.5 ha.

Cropland Time-series Dataset Reference

P. Potapov, S. Turubanova, M.C. Hansen, A. Tyukavina, V. Zalles, A. Khan, X.-P. Song, A. Pickens, Q. Shen, J. Cortez. (2021) Global maps of cropland extent and change show accelerated cropland expansion in the twenty-first century. Nature Food. https://doi.org/10.1038/s43016-021-00429-z 

Dataset Access

Global cropland data in GeoTiff and GEE formats available here: https://glad.umd.edu/dataset/croplands

View Data in Google Earth Engine App

Croplands: https://glad.earthengine.app/view/global-cropland-dynamics

Built-up Lands Extent and Change, 2000-2020

Built-up land consists of man-made land surfaces associated with infrastructure, commercial and residential land uses. At the Landsat spatial resolution, we define the built-up land class as pixels that include man-made surfaces, even if such surfaces do not dominate within the pixel. We employed the CNN (U-Net) algorithm calibrated with building outlines and road data from the Open Street Map to map the global built-up area extent for the years 2000 and 2020. Final 2000 and 2020 per-pixel class presence probabilities were generated, and validation data were used to assist in the final thresholding of the layers in depicting 2000 and 2020 extents and 2000-2020 gain in built-up lands. Loss of built-up lands, which represents a small proportion of the year 2000 class area, was not mapped by this provisional product.  

Global Surface Water Dynamics, 1999-2020

Open surface water, or simply water, is defined as inland water that covers ≥ 50% of a pixel and is not obscured by objects above the surface, for example, forest canopy, floating aquatic vegetation, bridges, or ice. Global maps derived from all Landsat 5, 7, and 8 scenes 1999-2020 highlight the changes in surface water extent during this period, and a probability sample-based assessment provides unbiased estimators of the area of permanent water, seasonal water, water loss, water gain, temporary land, temporary water, and high-frequency change 1999-2020. See detailed dataset description at the project webpage. 

Surface Water Dataset Reference

Pickens, A.H., Hansen, M.C., Hancher, M., Stehman, S.V., Tyukavina, A., Potapov, P., Marroquin, B., Sherani, Z., 2020. Mapping and sampling to characterize global inland water dynamics from 1999 to 2018 with full Landsat time-series. Remote Sensing of Environment 243, 111792. https://doi.org/10.1016/j.rse.2020.111792

Dataset Access

Global cropland data in GeoTiff and GEE formats available here: https://glad.umd.edu/dataset/global-surface-water-dynamics 

View Data in Google Earth Engine App

Water: https://glad.earthengine.app/view/surface-water-dynamics

Perennial Snow and Ice Extent and Change, 2000-2020

The perennial snow and ice layer includes land covered by glaciers and snow which remains during the entire year. The map is limited to the GLAD ARD tiles where perennial snow and ice areas were detected using the ARD observation quality layer. We mapped permanent snow and ice using regionally calibrated supervised classification models, one model per mountain system or a geographic region. Each classification model was calibrated with manually collected training data from multitemporal Landsat metrics. The models were applied to the years 2000 and 2020, and the results were filtered using a set of rules to remove noise in the time series.