Combining SAMOSA-3 and empirical retrackers for inland water height determination

Here we present a new system for retracking of SAR waveforms over rivers and lakes. Satellite altimetry offers frequent and global sampling across borders, which can be used to validate and calibrate hydrological models in remote areas where in situ measurements are scarce. The system was developed using CryoSat-2 20Hz SAR data, but due to the similarities between the Sentinel-3 SRAL altimeter and the SIRAL altimeter on-board CryoSat-2 an adaption of the method will be straightforward. The SAMOSA-3 retracker has previously demonstrated to perform better than other existing SAR retrackers over water surfaces and has therefore been chosen as the primary retracker whenever applicable. To find the waveforms for which the SAMOSA-3 retracker is appropriate, a classification is performed using first the k-means for clustering sample waveforms into classes and then the Naïve Bayes classifiers supported by training classes from the k-means for determining classes of all waveforms. The waveforms that are found suitable for the SAMOSA-3 retracker are isolated; these are the ocean-like and the highly specular waveforms. If the SAMOSA-3 retracker is found unsuitable or if the obtained result from this retracker is unsatisfactory, the waveform is retracked using the newly developed Multiple Waveform Persistent Peak (MWaPP) sub-waveform retracker, which studies adjacent waveforms to reveal the sub-waveform most likely to represent the echo from nadir. The method has been validated against lake and river gauges around the world to ensure that the system works for multiple types of inland water, and reveals that the combined method performs very well, but does not obtain significantly higher precisions compared to the MWaPP retracker. The results also show the MWaPP retracker provides height estimates with precisions very similar to those of the SAMOSA-3 retracker.