Accurate river water level estimation is critical for flood prevention, disaster mitigation, and water resource management, but complex plain river-network regions present severe challenges due to strong non-stationarity and nonlinearity. Traditional hydrological stations offer sparse spatial coverage and limited real-time capability, while satellite remote sensing alternatives each have drawbacks: conventional altimeters have sparse cross-track spacing, Synthetic Aperture Radar (SAR) has limited temporal resolution, and the Cyclone Global Navigation Satellite System (CYGNSS) suffers from coarse spatial resolution and observational noise. These limitations have hindered continuous monitoring in regions where water levels are jointly influenced by precipitation, sluice regulation, human activities, and agricultural irrigation.
In a study published in Satellite Navigation on October 12, 2026, researchers from Nanjing University of Posts and Telecommunications, Henan Polytechnic University, Anhui University, and the Nanjing Institute of Geography and Limnology of the Chinese Academy of Sciences introduced the Physics-Guided Noise-Enhanced Adaptive Fusion Dual-Stream Network (PGNAF-DSNet) for water level estimation over the complex river network of the Jiangsu plain in China. The model integrates Sentinel-1 SAR and CYGNSS observations through physics-guided feature construction, dual-stream temporal encoding, and adaptive feature fusion.
The study’s key innovations include physics-guided fused features—PFSWI, which multiplies Sentinel-1 backscatter differences by CYGNSS reflectivity, and Width_cv, the river width coefficient of variation—embedding physical priors into the model. A selective Gaussian noise enhancement strategy, optimized using the Tree-structured Parzen Estimator algorithm (σ = 0.04157), is applied only to channels susceptible to observational errors. The dual-stream architecture employs parallel Transformer and long short-term memory encoders to capture global and local temporal patterns, fused through an Adaptive Feature Fusion module that dynamically weights the two branches at the channel level.
On a strictly chronological test set (training: January 2021–November 2022; testing: May–October 2023), PGNAF-DSNet achieved a root mean square error of 0.2388 meters and a correlation coefficient of 0.9889, outperforming CNN, Transformer, LSTM, and Transformer-LSTM baselines. Notably, the model maintained R ≥ 0.95 across all forecast horizons up to T+6. Ablation experiments confirmed each component’s contribution, and SHAP analysis highlighted the high feature importance of Width_cv and PFSWI.
The authors stated that the study moves beyond simple data-level fusion, as previous approaches mainly used CYGNSS to fill temporal gaps in SAR observations without fully exploiting cross-modal feature interactions. By constructing physics-guided fused features and adopting a dual-stream architecture, the model captures both global trends and local dynamics of water level changes. The selective noise enhancement strategy acts as a regularizer, improving robustness against observational uncertainty and human-induced disturbances.
PGNAF-DSNet reduced prediction errors by approximately 50% in areas significantly affected by human activities such as sluice regulation and agricultural irrigation, demonstrating promise for operational forecasting. The framework could be extended to incorporate additional satellite observations, such as Surface Water and Ocean Topography water surface elevations, and adapted to other river systems through transfer learning. The full study is available via DOI 10.1186/s43020-026-00219-7.
