Jung et al. (2026) Towards global mapping of dynamic surface water extents using Sentinel-1 SAR data

Identification

• Journal: Remote Sensing of Environment
• Year: 2026
• Date: 2026-02-26
• Authors: Jungkyo Jung, Heresh Fattahi, Seongsu Jeong, Matthew Bonnema, John W. Jones, David Bekaert, S. Chan, Alexander L. Handwerger
• DOI: 10.1016/j.rse.2026.115326

Research Groups

• NASA Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA
• U.S. Geological Survey, Kearneysville, WV, USA

Short Summary

This paper introduces a fully automated and scalable method for mapping dynamic surface water extents from single-acquisition Sentinel-1 SAR imagery, integrating adaptive thresholding, fuzzy-logic classification, region growing, dark land estimation, and a bimodality test. The approach achieves classification accuracies exceeding 85% globally, providing a robust tool for near-real-time monitoring of floods, droughts, and water resources across diverse environmental conditions.

Objective

• To develop and assess a fully automated and scalable algorithm for delineating open surface water boundaries from single-acquisition Sentinel-1 SAR imagery, aiming for global applicability and high accuracy in monitoring temporal and spatial dynamics of water bodies.
• To specifically reduce false positives in low-backscattering areas and false negatives in high-backscattering water areas by incorporating a dual-polarization approach and advanced refinement steps.

Study Configuration

• Spatial Scale: Local to global domains, including 52 validation sites worldwide (approximately 2.5 km² each), near-global processing (1-12 November 2023), continental extent, and specific case studies in New York (USA), Los Angeles (USA), Upper Gulf of California (Mexico), Persian Gulf (Qatar), Grand Ethiopian Renaissance Dam (Ethiopia), Cerro Prieto Reservoir (Mexico), and Folsom Lake (California, USA).
• Temporal Scale: Single-acquisition Sentinel-1 images; validation using PlanetScope scenes from September–October 2019; global robustness evaluated with Sentinel-1 acquisitions from 1 to 12 November 2023; Grand Ethiopian Renaissance Dam monitored from June 2020 to January 2021; Cerro Prieto Reservoir monitored during summer seasons from 2016 to 2022; Folsom Lake monitored from May 2017 to September 2022.

Methodology and Data

• Models used:
  • Fully automated and scalable method integrating adaptive thresholding of radiometric terrain-corrected SAR backscatter data, fuzzy-logic classification, region growing, dark land estimation, and a bimodality test.
  • Kittler and Illingworth minimum error thresholding method.
  • Multi-level Otsu threshold algorithm (for trimodal distribution initialization).
  • Levenberg-Marquardt algorithm (for Gaussian curve fitting).
  • Total variation denoising (for speckle noise reduction).
  • Z- and S-shaped membership functions (for fuzzy logic).
  • Coefficient of variation (CV), Rx metric, Sarle's bimodality coefficient, Ashman D coefficient, and surface ratio (SR) for bimodality tests.
  • Height Above the Nearest Drainage (HAND) model for water boundary refinement.
• Data sources:
  • Primary Input: Sentinel-1 Interferometric Wide Swath (IW) product (C-band SAR, VV and VH polarizations), specifically OPERA Radiometric Terrain Corrected SAR Backscatter from Sentinel-1 (RTC-S1) product.
  • Ancillary Data: Copernicus digital elevation model (DEM) GLO-30, European Commission's Joint Research Centre (JRC) Global surface water occurrence map (1984–2021), European Space Agency (ESA) WorldCover data (2021).
  • Validation Data: PlanetScope optical images (3-meter resolution), OPERA Dynamic Surface Water Extent from Harmonized Landsat and Sentinel-2 (DSWx-HLS) products (30-meter resolution), National Agriculture Imagery Program (NAIP) aerial imagery, Maxar Worldview satellite imagery, Dynamic World land cover mapping database, USGS Landsat-based Dynamic Surface Water Extent (DSWE) product, Global Land Analysis & Discovery (GLAD) Monthly Water Product, and daily reservoir water-elevation records from the California Department of Water Resources.

Main Results

• The algorithm achieved classification accuracies exceeding 85% in detecting surface water extents across diverse environmental conditions.
• Validation with high-resolution optical imagery (52 global sites): Achieved an average overall accuracy of 93.27%, with a mean user's accuracy (UA) of 85.62% and a mean producer's accuracy (PA) of 93.55%.
• Near-global comparison with DSWx-HLS (6561 tiles): Yielded an overall UA of 94%, PA of 90%, and a Cohen's kappa coefficient (κ) of 0.94, confirming reliable performance at continental extent.
• Dark land suppression: In the Los Angeles case study, the combination of WorldCover masking and bimodality filtering significantly improved UA from 56.1% (region growing alone) to 99.84%, while maintaining PA above 98%.
• Grand Ethiopian Renaissance Dam (new construction): Successfully tracked rapid reservoir filling, with UAs between 95.6% and 98.54%, PAs from 82.3% to 91.8%, and κ from 0.9 to 0.95 against DSWx-HLS.
• Cerro Prieto Reservoir (drought): Accurately monitored drought-induced water loss from 2016 to 2022. Accuracies remained high until 2021 (PA > 85%, UA > 95%, κ > 0.95), with a decline in 2022 (PA 38.18%, UA 49.80%, κ 0.549) attributed to extremely low water levels and limited validation pixels.
• Folsom Lake (seasonal variation): Demonstrated capability to capture long-term seasonal water extent variations, showing strong linear correlations (R² values ranging from 0.93 to 0.98) between estimated water area and daily reservoir water-elevation records.

Contributions

• Introduces a novel, fully automated, and scalable methodology for global surface water extent mapping using single-acquisition Sentinel-1 SAR data, overcoming limitations of optical sensors (clouds, daylight).
• Develops a robust algorithm that effectively minimizes false positives in low-backscattering areas (e.g., sand, paved roads) and false negatives in high-backscattering water areas (e.g., wind-roughened water) through a sophisticated integration of adaptive thresholding, fuzzy-logic, region growing, dark land estimation, and bimodality tests, leveraging dual-polarization SAR data.
• Demonstrates high accuracy and reliability across a wide range of challenging environmental conditions, including arid regions, urban areas, new dam constructions, drought-stricken reservoirs, and seasonally varying lakes.
• Provides a practical and operational tool for near-real-time monitoring of floods, droughts, and water resource management at local to global scales.
• The developed methodology is sensor-agnostic, allowing for its potential application to forthcoming L- and S-band SAR missions like NISAR, broadening its applicability for future hydrologic observations.
• The algorithm is currently used to generate OPERA DSWx-S1 products, contributing to operational and scientific applications.

Funding

• Satellite Needs Working Group (SNWG) initiative
• National Aeronautics and Space Administration (NASA) through the Jet Propulsion Laboratory, California Institute of Technology

Citation

@article{Jung2026Towards,
  author = {Jung, Jungkyo and Fattahi, Heresh and Jeong, Seongsu and Bonnema, Matthew and Jones, John W. and Bekaert, David and Chan, S. and Handwerger, Alexander L.},
  title = {Towards global mapping of dynamic surface water extents using Sentinel-1 SAR data},
  journal = {Remote Sensing of Environment},
  year = {2026},
  doi = {10.1016/j.rse.2026.115326},
  url = {https://doi.org/10.1016/j.rse.2026.115326}
}