SWOT satellite: a more robust noise-reduction neural network for high-resolution mapping of ocean height

The U-Net model used to clean the KaRIn altimeter measurements produced artefacts in conditions of heavy swell or in the presence of internal waves. A new training procedure corrects these biases.

The French-American SWOT mission measures the height of the ocean surface with unprecedented resolution, revealing small-scale structures useful for the study of ocean circulation. But the two-dimensional observations of its KaRIn altimeter are marred by instrumental noise that masks precisely these high frequencies.

A U-Net-type convolutional network, proposed in 2023 to separate noise from the physical signal, had given good results in simulation and then on flight data. However, it showed weaknesses in atypical conditions, such as very strong swells or internal tidal solitons, sometimes with undesirable effects. The authors illustrate these limitations and present a modified version of the training, more robust, which avoids bias and artifacts in the denoised sea level. They also propose a standardised validation protocol, which can be used to compare other SWOT data filtering algorithms.