Fault tolerant approach to regenerate Level 1B SAR altimetry waveforms for enhancing Level 2 retrackers performance

This study deals with the identification and retrieval of anomalous waveforms generated in Level 1B processing chain of satellite altimetry over coastal areas and inland water bodies. Efficient identification of anomalous waveforms greatly improves the retracking performance, leading to the generation of precise water level time series that serve as vital inputs for hydrological studies. Abnormal behaviour in waveforms may be an indication of environmental changes, instrument malfunctions or other critical factors. To find anomalous waveforms, our framework utilizes an unsupervised machine learning technique. We categorise different parameters of the satellite's altimeter like AGC parameter, tracker range and features related to shape of waveforms for instance waveform’s skewness, number and location of peaks and so on for each sample in the dataset. Then we identify abnormal waveforms using a two-step density distribution probability analysis.

The secondary purpose of this study is proposing a robust strategy to retrieve abnormal waveforms in the level 1B SAR processing chain. This step is vital for narrow rivers and small inland water bodies, in which low number of measurements on related cycle cause missing hydrology data. In contrast to previous studies focusing solely on investigating L2 waveforms to determine precise retracking gates for multipeak and noisy waveforms, we propose an additional step in the L1B processing chain, specifically tailored to coastal and inland waters, enabling the retrieval of abnormal waveforms. In both fully focused and unfocused SAR processing, the final waveform is formed through the combination of various beam looks from the altimeter during fixed illumination time in stacks to the desired point on the surface, certain looks in the stack may exhibit undesirable patterns due to variations in environmental characterization, antenna footprint, and sidelobe gain. The proposed methods will mitigate the presence of undesirable waveforms in the stack prior to the generation of the final waveforms.

We apply the proposed methodology for Sentinel 3A and 3B datasets over different inland waters and validated our results against in-situ data. The results demonstrate that the water level time series, obtained by regenerated waveforms have significantly improved. The results show the potential of our proposed framework for detecting and retrieving anomalous waveforms leading to robust water level estimates from satellite altimetry data.