Testing of available land cover data for the territory of Armenia
The ESRI 2023 landcover data was selected as the basis for the project implementation. The ESA and GLAD landcover data can be additionally used for specific methodological tasks.
The choice was made based on the following reasons:
- GLC_FCS30D landcover data shows very strong excess of cropland area and excess of forest area (see here and here).
- Dynamic World data shows significant excess of cropland area, especially in the mountains as well as a number of other significant errors (see here and here).
- ESA, ESRI and GLAD data have fewer errors and are similar in their number and area. This is consistent with the results of comparing cropland area in landcover data with the ArmStat data, where the best match is shown by these three data sets.
- We have pre-tested ESA landcover data, but it does not allow us to test the methodology for assessing the dynamics of ecosystem area over a number of years. Therefore, ESRI data for 2023 was chosen as the main landcover for the project with the possibility of retrospective assessment of changes in ecosystem area since 2017.
- The possibilities of using the GLAD data for additional assessment of changes in ecosystem area and the degree of grassland degradation will be considered further in the course of the project.
The data for Armenia from the following publicly available global landcovers were tested:
1. Dynamic World
2. ESRI Land Cover
3. ESA WorldCover
4. GLC_FCS30D
5. GLAD Global Land Cover and Land Use Change.
Web-maps of tested land covers for the territory of Armenia →
Area of land cover classes in Armenia as a whole and in marzes →
Area of LC classes in landscape zones →
Comparison of cropland area according to ARMSTAT and land cover data →
Brief description of tested landcovers
Dynamic World
Primary link: https://dynamicworld.app/
Documentation: https://dynamicworld.app/about, https://www.nature.com/articles/s41597-022-01307-4
Where to get the data: via Google Earth Engine
Data provider: Google, World Resources Institute. License: Creative Commons BY-4.0.
Spatial resolution: 10 m
Temporal resolution: near real-time
Years of availability: 2015 – 2024.
Future availability: Project is based on two mature, well-known technologies: Google Earth Engine as processing and publishing engine and ESA Copernicus Sentinel-2 as data source. GEE is one of the key modern geospatial technologies. Sentinel-2 is a long-term program with scheduled activity up to 2033 (ref). These facts point to a secure future of Dynamic World.
Land cover classes:
1. Water
2. Trees
3. Grass
4. Flooded vegetation
5. Crops
6. Shrub and scrub
7. Built
8. Bare
9. Snow and ice
General commentary and issues:
Initially published in 2022, Google Earth Engine (GEE) based dynamic land cover dataset. Transparent and open-sourced. It is based on Sentinel-2 data and dynamically updated with new data acquisitions (3-5 days revisit time, excluding cloudy periods).
Could be challenging for inexperienced users to get data from GEE as files for analysis (designed to be used inside GEE).
Very basic classification scheme (e.g. single class “trees” for all forest types).
In general, there is no dataset in basic terms. There is a published machine learning algorithm which could be applied to any set of Sentinel-2 imagery, and this algorithm published together with the data at GEE. So users could request land cover data for particular territory based on a given period of Sentinel-2 acquisitions.
Python code sample to retrieve data from GEE (using geemap package): https://gist.github.com/eduard-kazakov/6bfa6ca1ab4ead0b2d6a3ed3e94dd277
ESRI Land Cover
Primary link: https://livingatlas.arcgis.com/landcover/
Documentation: https://www.impactobservatory.com/static/lulc_methodology_accuracy-ee742a0a389a85a0d4e7295941504ac2.pdf
Where to get the data: https://livingatlas.arcgis.com/landcoverexplorer
Data provider: ESRI. License: Creative Commons by Attribution (CC BY 4.0).
Spatial resolution: 10 m
Temporal resolution: 1 year
Years of availability: 2017 – 2023
Future availability: Land cover is provided by the world leader in geospatial, ESRI, and based on the well-known ESA Copernicus Sentinel-2 data. Sentinel-2 is a long-term program with scheduled activity up to 2033 (ref). These facts point to a secure future of ESRI Land Cover.
Land cover classes:
1. Water
2. Trees
3. Flooded vegetation
4. Crops
5. Built Area
6. Bare Ground
7. Snow/Ice
8. Clouds
9. Rangeland
General commentary and issues:
– Primary land cover product by ESRI, based on machine learning algorithms and Sentinel-2 data. Published every year. Available for direct download as GeoTIF for each year since 2017.
– Very basic classification scheme (e.g. single class “trees” for all forest types).
ESA WorldCover
Primary link: https://esa-worldcover.org/en
Documentation: https://worldcover2021.esa.int/documentation
Where to get the data: https://viewer.esa-worldcover.org/worldcover/
Data provider: ESA. License: Creative Commons Attribution 4.0 International
Spatial resolution: 10 m
Temporal resolution: 1 year
Years of availability: 2020 – 2021
Future availability: ESA has not officially confirmed that updates will follow annually, but the project has been extended due to its success and user demand. The current release patterns suggest that future updates might continue, though no fixed schedule has been guaranteed by ESA.
Land cover classes:
1. Tree cover
2. Shrubland
3. Grassland
4. Cropland
5. Built-up
6. Bare/sparse vegetation
7. Snow and Ice
8. Permanent water bodies
9. Herbaceous wetland
10. Mangroves
11. Moss and lichen
General commentary and issues:
– Flagman land cover project directed by ESA in cooperation with many partners. Based on Sentinel-2 and Sentinel-1 data (mixing optic and radar data). Distributed in GeoTIFF format via simple web interface.
GLC_FCS30D
Primary link: https://essd.copernicus.org/articles/16/1353/2024/
Documentation: https://essd.copernicus.org/articles/16/1353/2024/
Where to get the data: https://zenodo.org/records/8239305
Data provider: Liangyun Liu, Xiao Zhang, & Tingting Zhao. License: Creative Commons Attribution 4.0 International
Spatial resolution: 30 m
Temporal resolution: 1 year
Years of availability: 1985 – 2022
Future availability: Dataset is based on Landsat imagery. Three Landsat satellites are still active, the last one (Landsat 9) was launched in 2021. There are plans to continue the mission with Landsat Next in 2030/2031 (ref), so it seems that mission continuity is secure. According to latest publications, authors have intention to continue providing this data in the future. On the one hand they are supported and funded by the Chinese government, on the other hand the project obviously depended on particular scientists, which could be insecure.
Land cover classes:
1. Rainfed cropland
2. Herbaceous cover cropland
3. Tree or shrub cover (Orchard) cropland
4. Irrigated cropland
5. Open evergreen broadleaved forest
6. Closed evergreen broadleaved forest
7. Open deciduous broadleaved forest
8. Closed deciduous broadleaved forest
9. Open evergreen needle-leaved forest
10. Closed evergreen needle-leaved forest
11. Open deciduous needle-leaved forest
12. Closed deciduous needle-leaved forest
13. Open mixed leaf forest (broadleaved and needle-leaved)
14. Closed mixed leaf forest (broadleaved and needle-leaved)
15. Shrubland
16. Evergreen shrubland
17. Deciduous shrubland
18. Grassland
19. Lichens and mosses
20. Sparse vegetation
21. Sparse shrubland
22. Sparse herbaceous
23. Swamp
24. Marsh
25. Flooded flat
26. Saline
27. Mangrove
28. Salt marsh
29. Tidal flat
30. Impervious surfaces
31. Bare areas
32. Consolidated bare areas
33. Unconsolidated bare areas
34. Water body
35. Permanent ice and snow
General commentary and issues:
– This dataset is developed and supported by a group of scientists from different Chinese institutes. It’s well-known and cited hundreds of times, authors support it and add data for new years. Land cover is based on Landsat data time series. Project is supported by the National Natural Science Foundation of China.
– Product has a diverse classification scheme compared to other datasets.
– Data is distributed in zip archives available at famous scientific open data portal Zenodo, each GeoTIFF inside zip contains data for 20+ years (one band – one year).
GLAD Global Land Cover and Land Use Change
Primary link: https://glad.umd.edu/dataset/GLCLUC2020
Documentation: https://www.frontiersin.org/journals/remote-sensing/articles/10.3389/frsen.2022.856903/full
Where to get the data: https://storage.googleapis.com/earthenginepartners-hansen/GLCLU2000-2020/v2/download.html
Data provider: University of Maryland. License: Creative Commons Attribution 4.0 International
Spatial resolution: 30 m
Temporal resolution: 5 years
Years of availability: 2000 – 2020
Future availability: Dataset is based on Landsat imagery. Three Landsat satellites are still active, the last one (Landsat 9) was launched in 2021. There are plans to continue the mission with Landsat Next in 2030/2031 (ref), so it seems that mission continuity is secure. The GLAD project of University of Maryland is well-known and highly regarded by the community.
Land cover classes:
1. Terra Firma – True desert
2. Terra Firma – Semi-arid
3. Terra Firma – Dense short vegetation
4. Terra Firma – Tree cover
5. Wetland – Salt pan
6. Wetland – Sparse vegetation
7. Wetland – Dense short vegetation
8. Wetland – Tree cover
9. Open surface water
10. Snow/ice
11. Cropland
12. Built-up
13. Ocean
General commentary and issues:
– Well-known dataset by University of Maryland based on Landsat imagery archives. Project is focused on estimating global land use changes.
– Important property of this dataset is how it is detailed, with differentiation of trees by height, water retention time etc.
Datasets excluded from analysis
– MODIS MCD12Q1. We did not consider the MODIS data as a possible landcover for creating an ecosystem map due to its low resolution. However, these data can be used to assess ecosystem services.
Primary link:https://lpdaac.usgs.gov/products/mcd12q1v061/
Documentation:https://lpdaac.usgs.gov/documents/1409/MCD12_User_Guide_V61.pdf
Where to get the data: https://search.earthdata.nasa.gov/search
Data provider: NASA. License: No restrictions on reuse, redistribution, or modification
Spatial resolution: 500 m
Temporal resolution: 1 year
Years of availability: 2000 – 2023
Future availability: MCD12Q1 data is based on the MODIS sensor installed at Terra and Aqua satellites. According to the current plan, Terra MODIS will remain operational and generate the full suite of products until the end of the mission in December 2025, and Aqua MODIS will remain operational and generate the full suite of products until the end of the mission in August 2026 (ref). So we can await product availability up to 2025. This product will probably be replaced by a new generation one, but there is no particular information about it yet.
General commentary and issues: Well-known global Land Cover dataset, referenced thousands of times. Distributed with 8 different classification schemes. Training data haven’t been updated since 2021, so authors ask to be careful about data released after 2021 (ref). Relatively low spatial resolution.
– Copernicus Global Land Cover (https://land.copernicus.eu/en/products/global-dynamic-land-cover). Data is available only for 2015-2019, no further updates are planned. Other Copernicus products may be useful for assessing ecosystem services.
– ESA CCI/C3S Global Land Cover product (https://www.esa-landcover-cci.org/). Data is available only for 1992-2020. New releases were promised, but there were no actual updates in scheduled dates.
– Globeland30 (https://www.webmap.cn/commres.do?method=globeDetails&type=brief). Data is available only for 2000 and 2010, no further updates are planned.
– GlobCover (https://due.esrin.esa.int/page_globcover.php). Data is available only for 2009, no further updates are planned
– World Terrestrial Ecosystems (https://www.arcgis.com/home/item.html?id=926a206393ec40a590d8caf29ae9a93e). Data is available only for 2020, no further updates are planned
– The Global Land Cover by National Mapping Organizations (GLCNMO) (https://globalmaps.github.io/glcnmo.html). Data is available only for 2003-2013, no further updates are planned