Deep neural network based automatic grounding line delineation in DInSAR interferograms

Kurzfassung

The grounding line is a subsurface geophysical feature that divides a grounded ice sheet and floating ice shelf. Knowledge of its precise location is required for estimating ice sheet mass balance, as ice discharged from the interior is typically calculated at the grounding line [1]. Grounding lines move back and forth as ice shelves bend and flex due to ocean tides. Identifying their migration patterns can provide insights into understanding ice sheet dynamics and overall ice sheet stability [2] and thereby improve the accuracy of numerical ice sheet models. The spatial and temporal resolution of past and current satellite missions has enabled regular, continent-wide observation of Antarctica and other isolated glaciers with floating ice tongues. In particular, the high sensitivity of Interferometric SAR measurements to ground deformation has resulted in its application to grounding line location (GLL) mapping [3]. Specifically, the deformation at the grounding zone resulting from tidal flexure of the ice shelf is isolated from ice motion and topography in Differential InSAR (DInSAR) interferograms, under the assumption of steady ice velocity within the chosen temporal baseline. The tidal deformation is visible as a dense fringe belt and its landward limit is manually digitised as the GLL. Apart from being labour and time intensive, manual delineations are also inconsistent due to varying interpretations of experts in identifying the landward fringe, especially in areas with poor coherence or intricate fringe patterns. The concept of automatic GLL delineation has recently gained attention and seen the development of several methodologies. [4] demonstrated a semi-automatic method that estimates the fringe frequency of wrapped phase in DInSAR interferograms. The grounding zone can be directly identified by computing the gradient of the estimated frequencies, thereby avoiding phase unwrapping. However, this approach requires an a priori grounding zone location. [5] developed a deep learning based automatic delineation pipeline in which the proposed DNN was trained on real and imaginary components of DInSAR phases from Sentinel-1 acquisitions. This study further investigated the feasibility of DNNs for mapping the interferometric grounding line. The proposed DNN, based on the architecture of the Holistically-Nested Edge Detection network [6], was trained in a supervised manner, using manual delineations from the GLL product developed within ESA’s Antarctic Ice Sheet climate change initiative (AIS cci) project [7] as ground truth (Fig. 1 (a)). The GLL product contains manual delineations on 478 DInSAR interferograms computed from Sentinel-1A/B, ERS-1/2 and TerraSAR-X images acquired during 1992 - 2021. The training feature stack consists of four interferogram-based features: real and imaginary components, interferometric phase and pseudo coherence (which is estimated by applying a boxcar filter to interferometric phase) derived from the corresponding DInSAR interferograms and five auxiliary features derived from several compiled datasets: TanDEM-X Polar DEM [8], horizontal and vertical components of ice velocity [9], tidal amplitude [10] and atmospheric pressure [11]. An automatic workflow that handles the preparation of the training feature stack, training and inference of the neural network and the post processing of network generated delineations was developed. The performance of the neural network was evaluated as the median deviation of the network generated GLLs from the manual delineations, quantified using the PoLiS metric [12]. Additionally, the importance of individual features was indirectly gauged by training several networks with different feature subsets and comparing their median deviations from the ground truth. The DNN generated GLLs follow the landward limit of ice sheet flexure reasonably well, with the best network variant achieving a median deviation of 209 m from manual delineations.The contribution of auxiliary features was shown to be very weak, with their inclusion in the feature stack only slightly improving the delineation capability of the network. This finding is advantageous in terms of saving time, computational effort and memory in creating and storing the feature stack. References [1] E. Rignot and H. Thomas, “Mass balance of polar ice sheets,” Science, vol. 297, no. 5586, pp. 1502–1506, 2002. DOI: 10 . 1126 / science . 1073888. eprint: https : / / www . science . org / doi / pdf / 10 . 1126 / science.1073888. [Online]. Available: https://www.science.org/doi/abs/10.1126/science.1073888. [2] C. Schoof, “Ice sheet grounding line dynamics: Steady states, stability, and hysteresis,” Journal of Geophysical Research: Earth Surface, vol. 112, no. F3, 2007. [3] E. Rignot, “Tidal motion, ice velocity and melt rate of petermann gletscher, greenland, measured from radar interferometry,” Journal of Glaciology, vol. 42, no. 142, pp. 476–485, 1996. [4] A. Parizzi, “Potential of an Automatic Grounding Zone Characterization Using Wrapped InSAR Phase,” in IGARSS 2020 - 2020 IEEE International Geoscience and Remote Sensing Symposium, Waikoloa, HI, USA: IEEE, Sep. 2020, pp. 802–805, ISBN: 978-1-72816-374-1. DOI: 10.1109/IGARSS39084.2020.9323199. [5] Y. Mohajerani, S. Jeong, B. Scheuchl, I. Velicogna, E. Rignot, and P. Milillo, “Automatic delineation of glacier grounding lines in differential interferometric synthetic-aperture radar data using deep learning,” Scientific reports, vol. 11, no. 1, pp. 1–10, 2021. [6] S. Xie and Z. Tu, “Holistically-nested edge detection,” in Proceedings of the IEEE international conference on computer vision, 2015, pp. 1395–1403. [7] A. Groh, Product user guide (pug) for the antarctic ice sheet cci project of esa’s climate change initiative, version 1.0, 2021. [Online]. Available: https://climate.esa.int/media/documents/ST-UL-ESA-AISCCI-PUG-0001.pdf. [8] M. Huber, Tandem-x polardem product description, prepared by german remote sensing data center (dfd) and earth observation center, 2020. [Online]. Available: https://www.dlr.de/eoc/en/desktopdefault.aspx/tabid11882/20871_read-66374. [9] T. Nagler, H. Rott, M. Hetzenecker, J. Wuite, and P. Potin, “The sentinel-1 mission: New opportunities for ice sheet observations,” Remote Sensing, vol. 7, no. 7, pp. 9371–9389, 2015. [10] L. Padman, S. Erofeeva, and H. Fricker, “Improving antarctic tide models by assimilation of icesat laser altimetry over ice shelves,” Geophysical Research Letters, vol. 35, no. 22, 2008. [11] E. Kalnay, M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L. Gandin, M. Iredell, S. Saha, G. White, J. Woollen, et al., “The ncep/ncar 40-year reanalysis project,” Bulletin of the American meteorological Society, vol. 77, no. 3, pp. 437–472, 1996. [12] J. Avbelj, R. Muller, and R. Bamler, “A metric for polygon comparison and building extraction evaluation,” IEEE Geoscience and Remote Sensing Letters, vol. 12, no. 1, pp. 170–174, 2014.