Xu et al. (2025) Monitoring monthly dynamics of Nile River Basin surface water by combining Sentinel-1 SAR and Sentinel-2 multispectral imagery

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2025
• Date: 2025-10-16
• Authors: Jia Xu, Beibei Zhao, Vagner G. Ferreira, Ying Ge, Shaoguang Zhou, Hongyan Wang
• DOI: 10.1016/j.ejrh.2025.102860

Research Groups

• School of Earth Sciences and Engineering, Hohai University, Nanjing, China
• Land Satellite Remote Sensing Application Center, Ministry of Natural Resources of PR China, Beijing, China

Short Summary

This study developed a collaborative multi-source data integration framework (CMDIF) combining Sentinel-1 SAR and Sentinel-2 optical imagery to generate monthly surface water maps of the Nile River Basin (NRB) at 10-meter resolution. The approach achieved mapping accuracy exceeding 94.5%, offering improved spatiotemporal consistency and revealing distinct spatial and temporal patterns of surface water distribution across the NRB.

Objective

• To develop a Collaborative Multi-source Data Integration Framework (CMDIF) that synergistically integrates Sentinel-1 SAR and Sentinel-2 optical imagery to overcome single-sensor limitations in water body detection accuracy.
• To establish an automated sample generation methodology for consistent water extraction from both Sentinel-1 and Sentinel-2 datasets, enhancing classification robustness under complex environmental conditions.
• To generate a monthly seamless surface water dataset for the Nile River Basin (NRB) at 10-meter spatial resolution, addressing spatial discontinuity issues in existing products.
• To conduct a comprehensive quantitative assessment of the spatiotemporal dynamics of surface water variations across the NRB.

Study Configuration

• Spatial Scale: The Nile River Basin (NRB), covering approximately 3.4 million square kilometers across East and North Africa (longitudes 30°E to 40°E, latitudes 4°S to 31°N), subdivided into 10 sub-basins. Output resolution is 10 meters.
• Temporal Scale: Monthly dynamics from January 2019 to December 2021.

Methodology and Data

• Models used:
  • Collaborative Multi-source Data Integration Framework (CMDIF)
  • Google Earth Engine (GEE) platform for data acquisition and processing
  • Random Forest (RF) classifier for Sentinel-2 optical imagery
  • Isolation Forest (iForest) algorithm for outlier detection in training samples
  • Recursive Feature Elimination with Cross-Validation (RFE-CV) for feature selection
  • Simple Non-Iterative Clustering (SNIC) for superpixel segmentation of Sentinel-1 imagery
  • Edge-Otsu algorithm for water body extraction from Sentinel-1 imagery
  • Hierarchical decision tree for decision-level fusion of Sentinel-1 and Sentinel-2 results
• Data sources:
  • Satellite Imagery: Sentinel-1 SAR (dual-polarization VV, VH), Sentinel-2 MSI optical (B1-B12 bands).
  • Auxiliary Data:
    • Shuttle Radar Topography Mission (SRTM V3) for topographic information (elevation, slope, aspect).
    • OpenStreetMap and Global Georeferenced Database of Dams (GOODD) for dam locations.
    • Database for Hydrological Time Series of Inland Waters (DAHITI) for lake and reservoir water levels.
    • JRC GHSL human settlement layer for built-up area masking.
    • Land classification datasets: ESA Worldcover 2021, DynamicWorld, GLC_FCS30.
    • Water body datasets: JRC Global Surface Water (GSW), HydroShed, GRNWRZ V2.0.

Main Results

• Accuracy: Overall classification accuracy exceeded 94.5% (ranging from 94.67% to 96.78%), with Kappa coefficients greater than 0.89. Producer Accuracy and User Accuracy for surface water were both greater than 93%. Regional-scale validation showed mIoU > 82.52%, Recall > 83.51%, F1-score > 90.42%, and Precision > 94.16%.
• Surface Water Area: The average surface water area in the NRB from 2019 to 2021 was approximately 10.28 × 10^4 km^2. The largest inundation occurred in September 2020 (11.23 × 10^4 km^2), and the smallest in May 2019 (9.91 × 10^4 km^2).
• Spatial Distribution: Surface water is most concentrated in the upper reaches of the Nile (Victoria Nile, Lake Victoria, Lake Albert sub-basins), particularly between 3°S and 3°N. Lake Victoria sub-basin holds the highest proportion of surface water at 66.86% (6.95 × 10^4 km^2).
• Temporal Dynamics: A distinct seasonal pattern was observed, with surface water area increasing from June to September and decreasing from October to May each year.
• Water Body Types by Inundation Frequency: Permanent water bodies (WIF > 80%) accounted for 73.16%, seasonal water bodies (20% < WIF < 80%) for 4.79%, and temporary water bodies (5% < WIF < 20%) for 22.05%.
• Data Fusion Effectiveness: Sentinel-1 data showed a high participation rate in monthly surface water mapping, especially in cloud-prone regions like Lake Victoria, effectively supplementing optical data gaps and improving temporal continuity. The study's results showed a strong correlation (R^2 = 0.93) with DAHITI water level data for Lake Roseires.

Contributions

• Developed a novel Collaborative Multi-source Data Integration Framework (CMDIF) that effectively fuses Sentinel-1 SAR and Sentinel-2 optical imagery to overcome limitations of single-sensor approaches (e.g., cloud cover, data gaps) for continuous surface water monitoring.
• Introduced an automated training sample generation method, leveraging multi-source prior knowledge and the Isolation Forest algorithm for outlier detection, providing reliable and ample samples for large-scale basin-wide classification.
• Implemented a sub-region specific strategy for feature selection (RFE-CV) and independent Random Forest model training, enhancing classification accuracy and computational efficiency across the heterogeneous Nile River Basin.
• Integrated an object-based approach (SNIC superpixel segmentation with Edge-Otsu algorithm using SDWI) for Sentinel-1 data processing, effectively mitigating speckle noise and improving the detection of small and fragmented water bodies.
• Utilized a hierarchical decision tree for robust decision-level fusion of optical and SAR results, incorporating Dynamic World water occurrence frequency as a spatial prior and applying post-processing techniques to eliminate shadow interference.
• Generated a unique, monthly seamless surface water dataset for the entire Nile River Basin at 10-meter spatial resolution for 2019-2021, providing unprecedented detail on spatiotemporal water dynamics and addressing existing data discontinuity issues.
• The proposed framework offers high flexibility and practical value, making it adaptable for high spatiotemporal resolution surface water mapping in other large river basins globally, supporting sustainable water resource management and related scientific research.

Funding

• National Key R&D Program of China "Joint Research, Development and Application Demonstration of Remote Sensing Monitoring Technology for Typical Natural Resources Features" (Grant No. 2023YFE0207900).

Citation

@article{Xu2025Monitoring,
  author = {Xu, Jia and Zhao, Beibei and Ferreira, Vagner G. and Ge, Ying and Zhou, Shaoguang and Wang, Hongyan},
  title = {Monitoring monthly dynamics of Nile River Basin surface water by combining Sentinel-1 SAR and Sentinel-2 multispectral imagery},
  journal = {Journal of Hydrology Regional Studies},
  year = {2025},
  doi = {10.1016/j.ejrh.2025.102860},
  url = {https://doi.org/10.1016/j.ejrh.2025.102860}
}