Contents

Details

Analytics

Home

All Products

Docs

Source Cooperative is a Radiant Earth project

Global harmonized land cover datasets at 10m and 30m resolution (2020—2024) · OpenGeoHub Foundation · Source Cooperative | Source Cooperative

README

Global Harmonized Land Cover (GHLC)

Name: Global harmonized land cover datasets at 10m and 30m resolution (2020—2024)
Creator: OpenGeoHub Foundation

Data: Source Cooperative

This repository contains the processing and analysis workflow used to produce the Global Harmonized Land Cover (GHLC) dataset (Moreno et al., 2026). GHLC integrates multiple global and regional Earth observation products using a deterministic, rule-based ensemble approach to produce harmonized land cover maps at 10 m and 30 m spatial resolution. The product combines baseline land cover data inputs with specialized crop, grassland, forest, wetland, coastal, terrain, and other contextual layers to derive up to 44 detailed hierarchical land cover classes (see GHLC legend), and includes:

Product	Resolution	Temporal coverage
GHLC10	10 m	2020
GHLC30	30 m	2020 to 2024

The examples below compare the published 30 m and 10 m maps in 2020:

GHLC 10 m

GHLC 30 m

GHLC 10 m

GHLC 30 m

Workflow

The classification rules are ordered. Where multiple rules match a pixel, the first matching rule determines the final class. What you can find in this repository:

File / directory	Description
`config.py`	Input layers, class values, colors, temporal settings, and ordered land cover classification rules for the 10 m and 30 m workflows.
`lc_inputs_10m.csv`	Inventory of 10 m input data, including source information, temporal coverage, resolution, and resampling settings.
`lc_inputs_30m.csv`	Inventory of 30 m input data, including source information, temporal coverage, resolution, and resampling settings.
`landcover.md`	Main documented workflow for tiling, VRT preparation, land cover fusion, legends, and validation.
`preprocessing/01-landmask/`	Landmask preparation workflow.
`preprocessing/02-bare-soil-fraction/`	Bare soil fraction preprocessing from Landsat ARD inputs.
`lc_10m_lv1.qml`

The repository contains the code and method (explained in detail in Moreno et al., 2026). The published GHLC raster data are distributed separately through Source Cooperative.

GHLC legend

1. Built-up
1.1. Built-up	Description: Areas dominated by buildings, roads and other impervious, artificial surfaces. Assigned last in the rule chain, directly from the backbone built-up class, so every thematic rule takes precedence and only unclaimed pixels default to built-up. In the 30 m product, a reverse correction first reclassifies built-up pixels retaining >=30% tree cover with no recent forest loss as tree cover, addressing commission errors over street trees and urban parks. Key examples: Dense urban cores, residential blocks, industrial areas (e.g. Sao Paulo, Mumbai). Data inputs: ESA WorldCover (10 m only) GLAD GLCLUC (30 m only) Raster value: 1 (10 m & 30 m)
2. Cropland (annual and perennial)
2.1. Annual crops - soybean	Description: Annual cropland under soybean cultivation, assigned directly from the GLAD South America soybean layer. The crop-specific product is treated as authoritative within its domain, so the class is not gated by the backbone cropland class and is evaluated early in the chain, immediately after mangroves. Coverage is limited to South America, where the product is available (2000-2019). Key examples: Soybean fields in the Brazilian Cerrado and Argentine Pampas. Data inputs: GLAD Crop Map (soybean) Raster value: 2 (10 m & 30 m)
2.2. Annual crops - maize

Classes absent in the GHLC 30 m: Other grassland (3.3), Other shrubland (5.2), Moss and lichen (6.6), Wetland - other (7.4).

How to access the data

The public GHLC data are distributed through Source Cooperative. Use the repository page to browse and download available files.

Individual files can be accessed directly over HTTPS or anonymous S3:

Download with the AWS CLI (no credentials needed):

Since the files are Cloud-Optimized GeoTIFFs, GDAL, QGIS, and Rasterio can also stream directly over HTTPS without downloading the full global raster.

STAC access: S3 links are in each item's assets.

Interactive access: explore and summarize the data for an area of interest, without downloading the dataset available at LandMetric.

Example: plot an area of interest

The snippet below reads a small window from a GHLC Cloud-Optimized GeoTIFF (no full-file download needed), applies the official legend, and plots the result with a matching color legend. Edit the config block at the top to change the product, resolution, legend level, area, or title.

GHLC10 land cover, 2020 — Vietnam

Reproducing the workflow

The repository documents the scientific processing workflow, but the production scripts are currently configured for the OpenGeoHub production environment. Some scripts reference internal storage paths, network endpoints, tile lists, and object-storage aliases. These settings need to be adapted before running full global production on another system.

The workflow primarily uses Python and GDAL-based geospatial tooling. Some preprocessing components additionally use Bash, GNU Parallel, Julia, and S3-compatible object storage tools.

How to cite

The paper is currently in review. To cite this work, use:

Contact

For any questions or contributions, feel free to open an issue or reach out to author mateo.moreno@opengeohub.org.

Acknowledgements

Development of GHLC was supported by the Capitalizing on Earth Observation (CAPEO) project funded by the International Fund for Agricultural Development (IFAD) under project / grant number 2000005091.

This work was also supported by the Open-Earth-Monitor Cyberinfrastructure (OEMC) project, funded by the European Union's Horizon Europe research and innovation programme under grant agreement No. 101059548.

Disclaimer

The GHLC data and processing workflow are research products. OpenGeoHub Foundation makes no warranty, expressed or implied, regarding the accuracy, reliability, completeness, or fitness of these data for a particular purpose. Users are responsible for determining whether the data are suitable for their intended application and for the results of any use of the data.

Errors and artifacts may occur in the maps, source data, or processing workflow. If you identify an issue in the data or code, please report it through the repository issue. If an attribution or contributor has been unintentionally omitted, please contact OpenGeoHub Foundation so that it can be corrected in a future update.

License

This dataset is released under fully open license CC BY 4.0.

README

Global Harmonized Land Cover (GHLC)

Data: Source Cooperative

Product	Resolution	Temporal coverage
GHLC10	10 m	2020
GHLC30	30 m	2020 to 2024

The examples below compare the published 30 m and 10 m maps in 2020:

GHLC 10 m

GHLC 30 m

GHLC 10 m

GHLC 30 m

Workflow

The classification rules are ordered. Where multiple rules match a pixel, the first matching rule determines the final class. What you can find in this repository:

File / directory	Description
`config.py`	Input layers, class values, colors, temporal settings, and ordered land cover classification rules for the 10 m and 30 m workflows.
`lc_inputs_10m.csv`	Inventory of 10 m input data, including source information, temporal coverage, resolution, and resampling settings.
`lc_inputs_30m.csv`	Inventory of 30 m input data, including source information, temporal coverage, resolution, and resampling settings.
`landcover.md`	Main documented workflow for tiling, VRT preparation, land cover fusion, legends, and validation.
`preprocessing/01-landmask/`	Landmask preparation workflow.
`preprocessing/02-bare-soil-fraction/`	Bare soil fraction preprocessing from Landsat ARD inputs.
`lc_10m_lv1.qml`

The repository contains the code and method (explained in detail in Moreno et al., 2026). The published GHLC raster data are distributed separately through Source Cooperative.

GHLC legend

1. Built-up
1.1. Built-up	Description: Areas dominated by buildings, roads and other impervious, artificial surfaces. Assigned last in the rule chain, directly from the backbone built-up class, so every thematic rule takes precedence and only unclaimed pixels default to built-up. In the 30 m product, a reverse correction first reclassifies built-up pixels retaining >=30% tree cover with no recent forest loss as tree cover, addressing commission errors over street trees and urban parks. Key examples: Dense urban cores, residential blocks, industrial areas (e.g. Sao Paulo, Mumbai). Data inputs: ESA WorldCover (10 m only) GLAD GLCLUC (30 m only) Raster value: 1 (10 m & 30 m)
2. Cropland (annual and perennial)
2.1. Annual crops - soybean	Description: Annual cropland under soybean cultivation, assigned directly from the GLAD South America soybean layer. The crop-specific product is treated as authoritative within its domain, so the class is not gated by the backbone cropland class and is evaluated early in the chain, immediately after mangroves. Coverage is limited to South America, where the product is available (2000-2019). Key examples: Soybean fields in the Brazilian Cerrado and Argentine Pampas. Data inputs: GLAD Crop Map (soybean) Raster value: 2 (10 m & 30 m)
2.2. Annual crops - maize

Classes absent in the GHLC 30 m: Other grassland (3.3), Other shrubland (5.2), Moss and lichen (6.6), Wetland - other (7.4).

How to access the data

The public GHLC data are distributed through Source Cooperative. Use the repository page to browse and download available files.

Individual files can be accessed directly over HTTPS or anonymous S3:

Download with the AWS CLI (no credentials needed):

Since the files are Cloud-Optimized GeoTIFFs, GDAL, QGIS, and Rasterio can also stream directly over HTTPS without downloading the full global raster.

STAC access: S3 links are in each item's assets.

Interactive access: explore and summarize the data for an area of interest, without downloading the dataset available at LandMetric.

Example: plot an area of interest

GHLC10 land cover, 2020 — Vietnam

Reproducing the workflow

The workflow primarily uses Python and GDAL-based geospatial tooling. Some preprocessing components additionally use Bash, GNU Parallel, Julia, and S3-compatible object storage tools.

How to cite

The paper is currently in review. To cite this work, use:

Contact

For any questions or contributions, feel free to open an issue or reach out to author mateo.moreno@opengeohub.org.

Acknowledgements

Disclaimer

License

This dataset is released under fully open license CC BY 4.0.

1https://data.source.coop/opengeohub/ogh-ghlc10/<path-to-file>
2s3://us-west-2.opendata.source.coop/opengeohub/ogh-ghlc10/<path-to-file>

1@Article{Moreno2026GHLC,
2AUTHOR = {Moreno, M. and Parente, L. and Hein, L. and Bonannella, C. and Xu, Q. and Isik, M. S. and Wibisana, M. I. and Piccoli, M. and Hengl, T.},
3TITLE = {GHLC: global harmonized land cover datasets for 2020–2024 at 10 m and 30 m resolution based on a rule-based ensemble approach},
4JOURNAL = {EGUsphere},
5VOLUME = {2026},
6YEAR = {2026},
7PAGES = {1--33},
8URL = {https://egusphere.copernicus.org/preprints/2026/egusphere-2026-5540/},
9DOI = {10.5194/egusphere-2026-5540}
10}

1https://data.source.coop/opengeohub/ogh-ghlc10/<path-to-file>
2s3://us-west-2.opendata.source.coop/opengeohub/ogh-ghlc10/<path-to-file>

1@Article{Moreno2026GHLC,
2AUTHOR = {Moreno, M. and Parente, L. and Hein, L. and Bonannella, C. and Xu, Q. and Isik, M. S. and Wibisana, M. I. and Piccoli, M. and Hengl, T.},
3TITLE = {GHLC: global harmonized land cover datasets for 2020–2024 at 10 m and 30 m resolution based on a rule-based ensemble approach},
4JOURNAL = {EGUsphere},
5VOLUME = {2026},
6YEAR = {2026},
7PAGES = {1--33},
8URL = {https://egusphere.copernicus.org/preprints/2026/egusphere-2026-5540/},
9DOI = {10.5194/egusphere-2026-5540}
10}

1aws s3 cp s3://us-west-2.opendata.source.coop/opengeohub/ogh-ghlc10/<path-to-file> . --no-sign-request

1aws s3 cp s3://us-west-2.opendata.source.coop/opengeohub/ogh-ghlc10/<path-to-file> . --no-sign-request

1import requests, re
2import numpy as np
3import matplotlib.pyplot as plt
4from matplotlib.colors import ListedColormap, BoundaryNorm
5from matplotlib.patches import Patch
6import rasterio
7from rasterio.windows import from_bounds
8
9
10# 1. Configure: pick your map, legend, and area of interest here
11raster_url = "https://data.source.coop/opengeohub/ogh-ghlc10/land.cover_ifad.mcdt_c_10m_s_20200101_20201231_go_epsg.4326_v1.2.tif"
12qml_url = "https://data.source.coop/opengeohub/ogh-ghlc10/lc_10m_lv2.qml"
13bbox = (105.25, 10.50, 105.50, 10.75)  # (lon_min, lat_min, lon_max, lat_max)
14title = "GHLC10 land cover, 2020 — Vietnam"
15
16# 2. Load: parse the legend and read the raster window (no changes needed below)
17entries = re.findall(r'value="(\d+)".*?color="(#\w+)".*?label="([^"]*)"', requests.get(qml_url).text)
18legend = {int(v): (label, color) for v, color, label in entries}
19
20with rasterio.open(raster_url) as src:
21    data = src.read(1, window=from_bounds(*bbox, transform=src.transform))
22
23values = sorted(v for v in np.unique(data) if v in legend)
24cmap = ListedColormap([legend[v][1] for v in values])
25norm = BoundaryNorm(values + [values[-1] + 1], cmap.N)
26handles = [Patch(facecolor=legend[v][1], edgecolor="black", linewidth=0.3, label=legend[v][0]) for v in values]
27
28# 3. Plot
29fig, (ax_map, ax_leg) = plt.subplots(1, 2, figsize=(8, 5), gridspec_kw={'width_ratios': [3, 1]})
30ax_map.imshow(data, cmap=cmap, norm=norm)
31ax_map.set_title(title)
32ax_map.axis("off")
33ax_leg.axis("off")
34ax_leg.legend(handles=handles, loc="center left", frameon=False, fontsize=8)
35
36plt.subplots_adjust(wspace=0.02)
37fig.savefig("ghlc-example.png", dpi=200, bbox_inches="tight")
38plt.show()

1import requests, re
2import numpy as np
3import matplotlib.pyplot as plt
4from matplotlib.colors import ListedColormap, BoundaryNorm
5from matplotlib.patches import Patch
6import rasterio
7from rasterio.windows import from_bounds
8
9
10# 1. Configure: pick your map, legend, and area of interest here
11raster_url = "https://data.source.coop/opengeohub/ogh-ghlc10/land.cover_ifad.mcdt_c_10m_s_20200101_20201231_go_epsg.4326_v1.2.tif"
12qml_url = "https://data.source.coop/opengeohub/ogh-ghlc10/lc_10m_lv2.qml"
13bbox = (105.25, 10.50, 105.50, 10.75)  # (lon_min, lat_min, lon_max, lat_max)
14title = "GHLC10 land cover, 2020 — Vietnam"
15
16# 2. Load: parse the legend and read the raster window (no changes needed below)
17entries = re.findall(r'value="(\d+)".*?color="(#\w+)".*?label="([^"]*)"', requests.get(qml_url).text)
18legend = {int(v): (label, color) for v, color, label in entries}
19
20with rasterio.open(raster_url) as src:
21    data = src.read(1, window=from_bounds(*bbox, transform=src.transform))
22
23values = sorted(v for v in np.unique(data) if v in legend)
24cmap = ListedColormap([legend[v][1] for v in values])
25norm = BoundaryNorm(values + [values[-1] + 1], cmap.N)
26handles = [Patch(facecolor=legend[v][1], edgecolor="black", linewidth=0.3, label=legend[v][0]) for v in values]
27
28# 3. Plot
29fig, (ax_map, ax_leg) = plt.subplots(1, 2, figsize=(8, 5), gridspec_kw={'width_ratios': [3, 1]})
30ax_map.imshow(data, cmap=cmap, norm=norm)
31ax_map.set_title(title)
32ax_map.axis("off")
33ax_leg.axis("off")
34ax_leg.legend(handles=handles, loc="center left", frameon=False, fontsize=8)
35
36plt.subplots_adjust(wspace=0.02)
37fig.savefig("ghlc-example.png", dpi=200, bbox_inches="tight")
38plt.show()