Allred et al. (2025) Sentinel-2 based estimates of rangeland fractional cover and canopy gap class for the western United States

Identification

• Journal: Scientific Data
• Year: 2025
• Date: 2025-11-28
• Authors: Brady Allred, Sarah E. McCord, Timothy J. Assal, Brandon T. Bestelmeyer, Chad S. Boyd, Alexander C. Brooks, Samantha M. Cady, Michael C. Duniway, Samuel D. Fuhlendorf, Shane Green, Georgia R. Harrison, Eric R. Jensen, Emily Kachergis, Anna C. Knight, Chloe M. Mattilio, Brian A. Mealor, David E. Naugle, Dylan O'Leary, Peter J. Olsoy, Erika S Peirce, Jason R. Reinhardt, Robert K. Shriver, Joseph T. Smith, Jason D. Tack, Ashley M. Tanner, Evan P. Tanner, Dirac Twidwell, Nicholas P. Webb, Scott L. Morford
• DOI: 10.1038/s41597-025-06160-9

Research Groups

• Numerical Terradynamic Simulation Group, University of Montana, Missoula, MT, USA
• Jornada Experimental Range, USDA Agricultural Research Service, Las Cruces, NM, USA
• Bureau of Land Management, National Operations Center, Denver, CO, USA
• Eastern Oregon Agricultural Research Center, USDA Agricultural Research Service, Burns, OR, USA
• Desert Research Institute, Reno, NV, USA
• Department of Agronomy and Horticulture, University of Nebraska–Lincoln, Lincoln, NE, USA
• U.S. Geological Survey, Southwest Biological Science Center, Moab, UT, USA
• Natural Resource Ecology and Management, Oklahoma State University, Stillwater, OK, USA
• USDA Natural Resources Conservation Service, Central National Technology Support Center, Ft. Worth, TX, USA
• University of Wyoming Sheridan Research and Extension Center, Institute for Managing Annual Grasses Invading Natural Ecosystems, Sheridan, WY, USA
• W.A. Franke College of Forestry and Conservation, University of Montana, Missoula, MT, USA
• Institute for Natural Resources, Oregon State University, Corvallis, OR, USA
• Rangeland Resources and Systems Research Unit, USDA Agricultural Research Service, Fort Collins, CO, USA
• USDA Forest Service, Rocky Mountain Research Station, Moscow, ID, USA
• Department of Natural Resources and Environmental Science, University of Nevada, Reno, NV, USA
• U.S. Fish and Wildlife Service, Habitat and Population Evaluation Team, Missoula, MT, USA
• Caesar Kleberg Wildlife Research Institute, Texas A&M University-Kingsville, Kingsville, TX, USA

Short Summary

This paper develops and applies a temporal one-dimensional convolutional neural network using Sentinel-2 satellite data to produce annual, 10 m resolution estimates of rangeland fractional cover and canopy gap size classes for the western United States from 2018 to 2024, demonstrating improved accuracy over previous Landsat-based methods.

Objective

• To leverage Sentinel-2 satellite data to estimate fractional cover of common plant functional types, select genera, and canopy gap size classes across rangelands of the western United States.
• To produce annual, 10 m spatial resolution estimates of these rangeland characteristics for the years 2018 to 2024.

Study Configuration

• Spatial Scale: Rangelands of the 17 western states of the United States, distributed as 75 km × 75 km tiles. Spatial resolution of estimates is 10 m.
• Temporal Scale: Annual estimates for the years 2018 to 2024. Satellite data processed using 29 sequential 10-day timesteps per year.

Methodology and Data

• Models used: Temporal one-dimensional convolutional neural network (1D CNN) architecture.
• Data sources:
  • Satellite: Sentinel-2 (2A and 2B) top-of-atmosphere reflectance (10 m spatial resolution, 10-day nominal revisit time). Bands used: visible (2–4), near infrared (5–8), short-wave infrared (11-12). Derived indices: Normalized Difference Vegetation Index (NDVI), Normalized Burn Ratio 2 (NBR2).
  • Observation (Field Data): Line point intercept (LPI) and canopy gap measurements from various programs across the conterminous United States (47,833 samples from 2018-2024), aggregated and harmonized via Landscape Data Commons.
  • Ancillary Data: Cropland Data Layer (30 m resolution) for identifying non-rangeland areas (cultivated, developed, open water).
  • Processing Platform: Google Earth Engine.

Main Results

• Annual, 10 m spatial resolution estimates of fractional cover and canopy gap size class were produced for 2018-2024 across the 17 western states of the United States.
• Fractional cover estimates include annual forb and grass, bareground, litter, perennial forb and grass, shrub, tree, and select genera (invasive annual grass, pinyon-juniper, sagebrush).
• Canopy gap size classes include 25-50 cm, 51-100 cm, 101-200 cm, and greater than 200 cm.
• The model demonstrated good performance, with Root Mean Square Error (RMSE) for annual forb and grass, perennial forb and grass, shrub, and tree decreasing by 1.6%, 1.3%, 1.0%, and 0.4% respectively, compared to Landsat-based predictions (RAP v3).
• Mean Absolute Error (MAE) for the same groups decreased by 1.6%, 2.2%, 1.2%, and 0.3% respectively, compared to RAP v3.
• RMSE for canopy gap size classes 25-50 cm and greater than 200 cm decreased by 0.9% and 2.2% respectively, compared to available Landsat-based predictions.
• The 10 m Sentinel-2 based estimates showed improved detection of small features and better distinction of phenological signatures, aiding in evaluating management actions and identifying specific plant types like invasive annual grasses.

Contributions

• Development of a novel, high-resolution (10 m) dataset for rangeland fractional cover and canopy gap size classes for the western United States, leveraging the enhanced spatial, spectral, and temporal capabilities of Sentinel-2 satellites.
• Implementation of a temporal one-dimensional convolutional neural network for concurrent estimation of multiple rangeland vegetation characteristics, improving upon previous Landsat-based methods in terms of accuracy (lower RMSE and MAE).
• Provision of a publicly available, Cloud Optimized GeoTIFF dataset (approximately 2 TB) that offers more specific plant functional types and better captures spatiotemporal heterogeneity, enabling more precise rangeland monitoring and management decisions.

Funding

• SCINet project and/or the AI Center of Excellence of the USDA Agricultural Research Service, ARS project numbers 0201-88888-003-000D and 0201-88888-002-000D.
• Google Earth Engine.
• Numerical Terradynamic Simulation Group at the University of Montana.

Citation

@article{Allred2025Sentinel2,
  author = {Allred, Brady and McCord, Sarah E. and Assal, Timothy J. and Bestelmeyer, Brandon T. and Boyd, Chad S. and Brooks, Alexander C. and Cady, Samantha M. and Duniway, Michael C. and Fuhlendorf, Samuel D. and Green, Shane and Harrison, Georgia R. and Jensen, Eric R. and Kachergis, Emily and Knight, Anna C. and Mattilio, Chloe M. and Mealor, Brian A. and Naugle, David E. and O'Leary, Dylan and Olsoy, Peter J. and Peirce, Erika S and Reinhardt, Jason R. and Shriver, Robert K. and Smith, Joseph T. and Tack, Jason D. and Tanner, Ashley M. and Tanner, Evan P. and Twidwell, Dirac and Webb, Nicholas P. and Morford, Scott L.},
  title = {Sentinel-2 based estimates of rangeland fractional cover and canopy gap class for the western United States},
  journal = {Scientific Data},
  year = {2025},
  doi = {10.1038/s41597-025-06160-9},
  url = {https://doi.org/10.1038/s41597-025-06160-9}
}