Recovering noisy measurements over inland water bodies by regenerating L1B SAR altimetry waveforms using a segment-weighted Fully-Focused - Synthetic Aperture Radar (swFF-SAR) processing scheme

Satellite nadir altimetry has been a powerful technique for understanding oceans and seas over the past few decades. However, over rivers and small inland water bodies it produces noisy observations, which can result in gaps or erroneous measurements in water level time series. In this study, we aim to identify and correct anomalous measurements through reprocessing at the Level 1B (L1B) stage of the satellite altimetry processing chain.

To this end, we first detect abnormal waveforms that lead to anomalous water level measurements by analyzing various parameters related to the satellite's altimeter like AGC parameter and tracker range, and also waveform shape features. These waveform features include the number and location of peaks, noise level, kurtosis, centre of gravity, and peakiness. Abnormal waveforms are identified through an analysis of the distribution of these features.

While previous studies focused solely on L2 measurements to retrack multi-peak and noisy waveforms, we propose a robust strategy to regenerate abnormal waveforms within the L1B SAR processing chain by eliminating unwanted backscattered power. This approach incorporates the Fully-Focused Synthetic Aperture Radar technique into the L1B processing chain, dividing the illumination time into smaller stacks comprising multiple beam looks.

Due to factors such as antenna side lobe gain, wide antenna footprints, and environmental unevenness, some beam looks may exhibit undesired patterns. Our proposed approach addresses this issue by comparing the power of individual stacks with an analytically-derived reference waveform and assigning weights to each stack based on their similarity to the reference waveform. This reduces the impact of unwanted components in the final waveform and enables the regeneration of detected abnormal waveforms for inland waters.

We applied the proposed method to Sentinel-3A, Sentinel-3B, and Sentinel-6MF measurements over 6 lakes and reservoirs of various sizes and validated the results against in-situ data. The validation demonstrates that the water height time series obtained from regenerated waveforms match significantly better with in-situ measurements. Specifically, the accuracy of the water level time series, measured in terms of RMSE, improved by around 60% for the selected case studies after applying retracking on newly generated waveforms.