X-band COSMO-SkyMed Data Grounding Line Mapping Using Deep Learning

The grounding line marks the transition between the floating and grounded part of a glacier. As a sensitive indicator of sea level changes, grounding lines can be monitored using the differential interferometric synthetic aperture radar (DInSAR) technique [1]. Accurate measurements of grounding line migrations are also necessary to correctly asses the retreat of a glacier. While state of the art grounding line mapping methodologies requires a human expert in the loop, deep learning approaches have been implemented for automatic grounding line mapping at C-band using the ESA Sentinel-1A/B constellation [2]. However, it has been shown that the 6 days C-band (5.6 cm wavelength) interferometry is sub-optimal for grounding line mapping as there is phase decorrelation over the grounding zones of fast-moving Antarctic glaciers [3]. On the contrary, X-band (3 cm wavelength) COSMO-SkyMed 1-day interferometry proved to be the most effective existing configuration to map rapidly changing glaciers. In this paper, we train a deep learning neural network to map grounding lines on COSMO-SkyMed X-band data. The neural network we are using in this research was originally successfully implemented by Y. Mohajerani et. al. [2] to treat C-band DInSAR data. Here we show how the proposed approach can be extended to effectively map grounding lines on X-band DInSAR data, acquired by the ASI's COSMO-SkyMed mission in 2020 - 2022 over Antarctica. We compare the results of manual and automatic mapping processes and show that there are situations where the neural network manages to map grounding lines where a human expert cannot.

Fig 1: Phase tile used for the neural network testing, corresponding manually mapped grounding line, and network's output for the different number of training epochs.

References:

[1] Rignot, E., et al. (2014). Widespread, rapid grounding line retreat of Pine Island, Thwaites, Smith, and Kohler glaciers, West Antarctica, from 1992 to 2011. Geophysical Research Letters, 41, 3502-3509.

[2] Mohajerani, Y., et al. (2021). Automatic delineation of glacier grounding lines in differential interferometric synthetic-aperture radar data using deep learning. Scientific reports, 11, 1-10.

[3] Milillo, P., et al. (2019). Heterogeneous retreat and ice melt of Thwaites Glacier, West Antarctica. Science advances, 5, eaau3433.