Abstract's details

Detecting sub-waveforms using spatio-temporal waveform information in combination with sparse representation and conditional random fields for coastal retracking applications

Bernd Uebbing (University of Bonn, Institute for Geodesy and Geoinformation, Germany)


Ribana Roscher (University of Bonn, Institute for Geodesy and Geoinformation, Germany); Jürgen Kusche (University of Bonn, Institute for Geodesy and Geoinformation, Germany)

Event: 2016 Ocean Surface Topography Science Team Meeting

Session: Instrument Processing: Measurement and retracking (SAR and LRM)

Presentation type: Type Oral

Contribution: PDF file


For more than two decades, satellite radar altimeters have been providing valuable information on variations of sea surface height (SSH) of seas and oceans. Unfortunately, reliable observations in coastal waters are scarce due to strong influences from land returns inside the altimter footprint on the measured waveform. Recently, methods have been introduced to mitigate the land effect, e.g., either by utilizing a two step approach where the significant waveheight is fixed during the second iteration (Passaro et al., 2014), or introducing a pre-processing step to derive sub-waveforms and applying a threshold method to individual sub-waveforms, which requires additional prior information on the SSH (Hwang et al., 2006). While both methods allow to retrieve SSHs closer to the coast compared to conventional retracking approaches, they still produce a number of outliers.

We present a novel approach for deriving sub-waveforms using spatio-temporal information between neighboring range gates within one waveform and between continuous waveforms by employing a sparse representation approach which is modeled as a conditional random field (CRF). Then, sub-waveforms are defined as coherent range gates within one waveform that are modeled by the same set of base elements. Our sub-waveform detection divides the whole waveforms into sub-waveforms where we can select a rather coarse or fine division depending on pre-defined hyperparameters of the CRF. In contrast to the method proposed by Hwang et al. (2006) where only possible leading edges were detected, we divide the whole waveform which gives us potential leading edges, as well as other distinct shape features.
Here, we combine the derived sub-waveforms with a 3-parameter ocean retracker, either directly by fitting only to the individual sub-waveforms of the whole waveform or by using the sub-waveform division to derive a weighting scheme for the total waveform to be employed during the parameter estimation. At first, we compute a pointcloud with more than one SSH at each 20Hz position. Then, the pointcloud is further processed by applying a Dijkstra algorithm to detect the best SSHs at each position, resulting in 20Hz retracked SSHs.

We employ our approach to coastal waveforms from Envisat and Jason-2 in the Mediterranean Sea, as well as in the Bay of Bengal. Our results show that over the open ocean the proposed method produces SSHs that are very similar to ones derived from conventional ocean retracking algorithms, while in coastal regions we are able to derive meaningful SSHs up to less than 1km off the coast of at least the same quality, and for a larger number of available cycles compared to existing methods. For validation of the coastal estimates, we compare to high temporal resolution tide gauge data from tide gauges located in Triest (Italy), as well as Chittagong (Bangladesh).

Oral presentation show times:

Room Start Date End Date
Auditorium Tue, Nov 01 2016,17:30 Tue, Nov 01 2016,17:45
Bernd Uebbing
University of Bonn, Institute for Geodesy and Geoinformation