glance
The NASA-funded Global Land Cover Estimation (GLanCE) project seeks to provide high-quality long-term records of land cover and land cover change at a 30 m spatial resolution for the 21st century (2001 to present) (1). The GLanCE training dataset was designed for regional-to-global land cover and land cover change analyses (2, 3). Similar to LCMAP, the dataset legend is general-purpose and intended to support a broader community of end-users; however, the GLaNCE dataset has a global (as opposed to national) scope.
Our glance evaluation dataset is derived from the GLanCE training dataset in
the GEE Community Catalog
("projects/sat-io/open-datasets/GLANCE/GLANCE_TRAINING_DATA_V1") (4). Though
published GLaNCE data products use Level 1 labels (5), we use the Level 2 of the
labeling hierarchy ("Glance_Class_ID_level2") as a test of maximizing thematic
detail. Given that GLaNCE includes a number of other datasets, some of which
overlap with other evaluation datasets, e.g., LCMAP, we select a subset of
sources, specifically the MODIS STEP dataset (STEP), results of
spectral-temporal clustering (CLUSTERING), a labeled dataset from the NASA
Arctic-Boreal Vulnerability Experiment (ABoVE), and a set of annotations
collection by the project team ("Dataset_Code" = 1, 2, 4, or 704).
GLaNCE labels are associated with time segments, i.e., labels have a start and
end date similar to our use of a valid period. We select only labeled segments
with an end year after 2017 ("End_Year" greater than or equal to 2017). We
remove null values as well as the "ice_and_snow" and "moss" categories,
which have fewer than 500 samples per class. This results in a final dataset
with eleven classes, and we select 300 training points per class with the
remainder allocated to the test split. Though we note that segments could be
converted to a series of annual labels for each location, this approach would be
subject to greater temporal autocorrelation across labels for the same location;
instead, we sample a random year between segment start and end dates as an
annual valid period to ensure more independent sampling of the time domain. Our
final GLaNCE land cover evaluation has a total of 3,300 training points and
31,585 test points after pre-processing and spatial proximity filtering.
| label | label_name | train_count | test_count |
|---|---|---|---|
| 0 | water | 300 | 1167 |
| 1 | developed | 300 | 878 |
| 2 | soil | 300 | 291 |
| 3 | rock | 300 | 921 |
| 4 | sand | 300 | 1195 |
| 5 | deciduous | 300 | 2700 |
| 6 | evergreen | 300 | 5959 |
| 7 | mixed | 300 | 2425 |
| 8 | shrub | 300 | 2118 |
| 9 | grassland | 300 | 6952 |
| 10 | agriculture | 300 | 6979 |
License
GLanCE: A Global Land Cover Training Dataset from 1984 to 2020, from Boston University Global Land Cover Estimation (GLanCE), is licensed under the Creative Commons Attribution 4.0 International License (CC-BY). You may obtain a copy of the CC-BY license at: https://creativecommons.org/licenses/by/4.0/legalcode. You can obtain a copy of the dataset at https://source.coop/repositories/boston-university/bu-glance/access and/or https://gee-community-catalog.org/projects/glance_training/?h=glance. This version of the dataset is modified as described above.
For the dataset citation, please see (3) in the “References” section below.
References
- M. A. Friedl, C. E. Woodcock, P. Olofsson, Z. Zhu, T. Loveland, R. Stanimirova, P. Arevalo, E. Bullock, K.-T. Hu, Y. Zhang, K. Turlej, K. Tarrio, K. McAvoy, N. Gorelick, J. A. Wang, C. P. Barber, C. Souza Jr, Medium spatial resolution mapping of global land cover and land cover change across multiple decades from Landsat. Front. Remote Sens. 3 (2022).
- R. Stanimirova, K. Tarrio, K. Turlej, K. McAvoy, S. Stonebrook, K.-T. Hu, P. Arévalo, E. L. Bullock, Y. Zhang, C. E. Woodcock, P. Olofsson, Z. Zhu, C. P. Barber, C. M. Souza Jr, S. Chen, J. A. Wang, F. Mensah, M. Calderón-Loor, M. Hadjikakou, B. A. Bryan, J. Graesser, D. L. Beyene, B. Mutasha, S. Siame, A. Siampale, M. A. Friedl, A global land cover training dataset from 1984 to 2020. Sci. Data 10, 879 (2023).
- R. Stanimirova, K. Tarrio, K. Turlej, K. McAvoy, S. Stonebrook, K.-T. Hu, P. Arévalo, E. L. Bullock, Y. Zhang, C. E. Woodcock, P. Olofsson, Z. Zhu, C. P. Barber, C. M. Souza Jr, S. Chen, J. A. Wang, F. Mensah, M. Calderón-Loor, M. Hadjikakou, B. A. Bryan, J. Graesser, D. L. Beyene, B. Mutasha, S. Siame, A. Siampale, M. A. Friedl, "A Global Land Cover Training Dataset from 1984 to 2020", Version 1.0, Radiant MLHub (2023); https://doi.org/10.34911/rdnt.x4xfh3.
- S. Roy, T. Swetnam, A. Saah, Samapriya/awesome-Gee-Community-Datasets: Community Catalog, Zenodo (2025); http://dx.doi.org/10.5281/ZENODO.14757583.
- P. Arevalo, R. Stanimirova, E. Bullock, Y. Zhang, K. Tarrio, K. Turlej, K. Hu, K. McAvoy, V. Pasquarella, C. Woodcock, P. Olofsson, Z. Zhu, N. Gorelick, T. Loveland, C. Barber, M. Friedl, Global Land Cover Mapping and Estimation Yearly 30 m V001. NASA EOSDIS Land Processes Distributed Active Archive Center (2022); https://lpdaac.usgs.gov/products/glance30v001/.