A physics-based component-wise simulation framework decomposes Sentinel-3 radar altimetry waveforms into river and interference contributions using topography and water masks. Across 55 validation virtual stations, it reduced median unbiased root-mean-square error (ubRMSE) to 0.47 m, outperforming standard retrackers and recovering water levels at complex reaches where operational databases routinely discard observations.
Satellite radar altimetry monitors inland water, but narrow rivers remain difficult. Sentinel-3 SAR mode improves along-track resolution to about 300 m, yet its pulse-limited across-track footprint can span kilometers. In floodplains, the footprint illuminates the target river, land and off-nadir lakes or wetlands simultaneously. Calm lateral water bodies can produce specular returns that compete with the nadir river echo, causing snagging and range biases. Empirical and analytical retrackers often treat aggregated waveforms statistically or assume uniform scattering, so they cannot attribute peaks to specific surfaces. Based on these challenges, in-depth research is needed on physics-based decomposition of altimetry echoes over heterogeneous rivers.
Published (DOI: 10.34133/remotesensing.1075) on 11 September 2026 in J ournal of R emote S ensing , this study by researchers at the Southern University of Science and Technology, China, introduces a physics-based component-wise simulation (PhysCompSim) framework. This Sentinel-3 SAR altimetry study, with implications for Sentinel-6, addresses a persistent problem in global satellite hydrology: off-nadir water bodies contaminate nadir echoes, causing tracking failures and data loss over heterogeneous rivers. By modeling how individual water patches contribute to the received waveform, the team aims to retrieve continuous water levels where conventional processors struggle.
The core advance is a bottom-up, physics-based decomposition. Rather than fitting the final waveform statistically, PhysCompSim simulates separate return components from the target river and each interfering water body, then fits their sum to observations. Across 55 validation stations, it achieved a median unbiased root-mean-square error (ubRMSE) of 0.47 m, compared with 0.79 m for the conventional Offset Center of Gravity (OCOG) retracker, 0.71 m for the Synthetic Aperture Radar Altimeter MOde Studies and Applications (SAMOSA) model, and 0.59 m for modified Multiple Waveform Persistent Peak (MWaPP+). That amounts to accuracy gains of roughly 20–41% over the three benchmarks. At eight stations with no standard product in either the Database for Hydrological Time Series over Inland Waters (DAHITI) or Hydroweb, the method still reconstructed continuous time series, showing recovery in severely contaminated reaches.
The framework combines a 30 m forest- and building-corrected elevation model (FABDEM) with dynamic Global Surface Water Occurrence masks. It discretizes the footprint into triangular facets, labels the target river and spatially disjoint interference bodies, and simulates delay-Doppler waveforms using antenna gain, range, Doppler stacking and surface roughness. A separable nonlinear least-squares solver separates the time shift and surface-roughness terms from linear amplitudes and noise. Of 25,213 waveforms, 30.8% were multipeak. Simulated and observed echoes matched closely, with median correlations of 0.97 for single-peak and 0.93 for multipeak cases; median normalized RMSE was 2.49% and 5.75%, respectively. Synthetic tests showed that off-nadir distance, elevation, area, orientation and roughness govern interference timing and strength. In a 26-station benchmark, mean ubRMSE was 0.46 m, versus 0.49 m for DAHITI and 0.54 m for Hydroweb. The model thus turns multipeak retracking into a constrained source-attribution task.
The authors state that the framework shifts retracking from blind waveform fitting to constrained source attribution. They note that isolating a radiometrically weaker river component enables retrieval in entangled settings and clearly positions PhysCompSim as a specialized, high-precision complement to operational retrackers, with potential for future Sentinel-3 and Sentinel-6 missions.
The method combines satellite state vectors, FABDEM terrain and water masks to reconstruct the radar observation geometry in a local reference frame. The scene is meshed into triangular facets and semantically segmented into river, interference and land classes. Backscatter contributions are simulated from each class and summed under incoherent scattering. The model is fitted to 20-Hz Sentinel-3A/B Level-2 SAR waveforms from 2016–2024. Water surface elevation is derived after geophysical and geoid corrections and validated against in situ daily mean gauge records.
Future work will target narrower rivers by coupling PhysCompSim with fully focused SAR processing and will incorporate time-varying topography to handle channel migration. The team also sees potential for multitarget elevation monitoring, such as retrieving adjacent wetlands or off-nadir channels from one overpass. With further automation and GPU acceleration, the approach could complement global databases and support more reliable basin-scale monitoring, flood preparedness and sustainable water resource management, especially where conventional altimetry discards contaminated observations from Sentinel-3 and Sentinel-6 missions.
###
References
DOI
Original Source URL
https://spj.science.org/doi/10.34133/remotesensing.1075
Funding information
This work was supported by the National Natural Science Foundation of China (No. 42471348), the High-level University Special Fund (No. G03050K001), and the Shenzhen Science and Technology Program (No. KCXFZ20240903093659003).
About J ournal of R emote S ensing
J ournal of R emote S ensing , an online-only Open Access journal published in association with AIR-CAS, promotes the theory, science, and technology of remote sensing, as well as interdisciplinary research within earth and information science.
Journal of Remote Sensing
Not applicable
A Novel Approach to Retrieve River Water Levels by Decomposing Altimetry Waveforms
11-Sep-2026
The authors declare that they have no competing interests.