CoAuthors

Mohammad J. Tourian (Institute of Geodesy, University of Stuttgart, Germany)

Single-waveform retracking for satellite altimetry applications of inland waters has reached its limits, obtaining decimeter-level accuracy or worse. The existing retracking methods find a retracker offset in a waveform by analyzing the variation in power along with the bin coordinate. This makes the retracking procedure strongly dependent on noise. Moreover, the success of such methods is only guaranteed for certain waveform types requiring cumbersome pre-processing steps, including waveform classification.
In this study, we collect neighboring waveforms into a radargram. The so-called radargram contains, unlike single waveforms, information on spatial variation of backscattered power over water surfaces. The radargram eases the recognition of patterns like retracking gate, off-nadir pattern (e.g., parabola), shoreline, etc. Instead of a retracking gate as a point in the 1D waveform, in a 2D radargram a line (referred to as a retracking line) is to be determined. In fact, by finding a retracker line in a radargram, each radargram can be segmented into two parts: the left and right hand side of the retracking line. This can be interpreted as a binary image segmentation problem in a more straightforward representation, in which spatial constraints are considered.
We formulated this problem using Markov Random Fields (MRF), which explicitly model the interaction between different constraints and auxiliary sources of information in a radargram. In such a formulation, we deal with a Bayesian framework with the goal of finding a specific labeling structure of the image which maximizes the posterior estimation of the MRF (MAP-MRF).
We evaluate our method using Jason-2 satellite altimetry data over 6 lakes in the Mississippi River basin and validate our results against in situ data. Validation shows the benefits of retracking radargrams instead of single-waveforms.