Andros et al. (2026) Swamp-Eye: a deep learning model for monitoring wetlands change across the globe

Identification

• Journal: Scientific Reports
• Year: 2026
• Authors: Charles Andros, Ian Conery, David S. Vinson, Katherine DeVore, Tristan D. Calaway, André Rovai, Jin Ikeda, Adam Collins, Yoko Masue-Slowey
• DOI: 10.1038/s41598-026-39257-1

Research Groups

• Environmental Laboratory, U.S. Army Engineer Research & Development Center (USACE-ERDC), Vicksburg, USA.
• Coastal and Hydraulics Laboratory, U.S. Army Engineer Research & Development Center (USACE-ERDC), Vicksburg, USA.
• Smithsonian Environmental Research Center, Edgewater, USA.
• Center for Computation and Technology, Louisiana State University, Baton Rouge, USA.

Short Summary

This study develops "Swamp-Eye," a deep learning model designed for the rapid and cost-effective monitoring of global wetland extent changes. By training on a diverse, multi-seasonal dataset of coastal and inland systems, the model achieved a 93.7% overall accuracy, providing a generalizable tool for large-scale environmental assessment.

Objective

• To create a generalizable deep learning framework capable of identifying and monitoring changes in wetland extent across diverse global geographies and seasonal conditions, particularly for remote or inaccessible regions.

Study Configuration

• Spatial Scale: Global (including diverse coastal and inland wetland systems across multiple continents).
• Temporal Scale: Multi-seasonal (utilizing imagery acquired throughout the year to account for seasonal variations in hydrology and vegetation).

Methodology and Data

• Models used: 15 candidate deep learning (DL) models were evaluated, including architectures based on semantic segmentation (referencing U-Net and DeepLabV3+ frameworks).
• Data sources: Harmonized Sentinel-2 MSI (MultiSpectral Instrument) Level-2A Surface Reflectance data; Global Mangrove Watch (GMW); Global Lakes and Wetlands Database (GLWD); and Google Earth Engine (GEE) for planetary-scale geospatial analysis.

Main Results

• The "Swamp-Eye" model emerged as the top performer with an averaged overall accuracy of 93.7%.
• Quantitative performance metrics included a producer’s accuracy of 79.4%, a user’s accuracy of 93.2%, and an Intersection over Union (IoU) of 74.6% across test sites.
• The model demonstrated high robustness across varying wetland types, including both coastal marshes and inland systems, despite seasonal fluctuations.

Contributions

• Introduction of a unique annotation system that integrates multiple global datasets to create a comprehensive, annotated imagery database for wetland classification.
• Development of a highly generalizable deep learning tool that addresses the limitations of site-specific models, enabling consistent global monitoring.
• Public release of the "Global Semantic Annotation Database" (GSADB), including code and imagery, to facilitate open-source advancement in remote sensing of wetlands.

Funding

• US Army Congressional Add line-item number PE 0603119 A, entitled "Ground Advanced Technology - Engineering Practices for Ecosystem Design Solutions."

Citation

@article{Andros2026SwampEye,
  author = {Andros, Charles and Conery, Ian and Vinson, David S. and DeVore, Katherine and Calaway, Tristan D. and Rovai, André and Ikeda, Jin and Collins, Adam and Masue-Slowey, Yoko},
  title = {Swamp-Eye: a deep learning model for monitoring wetlands change across the globe},
  journal = {Scientific Reports},
  year = {2026},
  doi = {10.1038/s41598-026-39257-1},
  url = {https://doi.org/10.1038/s41598-026-39257-1}
}

Generated by BiblioAssistant using gemini-3-flash-preview (Google API)