EGU25-17824, updated on 15 Mar 2025
https://doi.org/10.5194/egusphere-egu25-17824
EGU General Assembly 2025
© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.
Autoregressive mark-tracing for radiostratigraphy: A lightweight model for annotating internal reflection horizons in ice sheets
Hameed Moqadam1,2,3, Troels Arnfred Bojesen2, and Olaf Eisen1,4
Hameed Moqadam et al.
  • 1Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research, Bremerhaven, Germany (hameed.moqadam@awi.de)
  • 2Department of Informatics, University of Bergen, Bergen 5007, Norway
  • 3Constructor University, Bremen, Germany
  • 4University of Bremen, Bremen, Germany

Tracing internal reflection horizons (IRHs) in radio-echo sounding data is crucial for understanding ice sheet dynamics and reconstructing past climate conditions. We present an autoregressive generative model designed to trace IRHs iteratively, mimicking the human annotation process. Unlike conventional segmentation-based approaches, which require large training datasets and yield one-shot predictions necessitating extensive post-processing (Moqadam et al. 2024), our model works by estimating a spatial probability map for each annotation mark, conditioned on previously generated marks. This iterative approach emulates human-like tracing by sequentially traversing along each IRH and allows the model to learn from minimal data, resulting in transferability to diverse radar systems.

The model produces interpretable probability maps at each step, providing transparent outputs that human experts can verify directly, without the need for post hoc analyses. Furthermore, avoiding explicit class definitions mitigates the detrimental effects of imbalanced data, which is a common issue in traditional pixel classification methods. The lightweight design of the model – an iterative rather than one-shot approach – improves its suitability for widespread application. This innovative approach presents a significant advancement in automating the annotation of IRHs and provides a robust, interpretable, and adaptable solution for ice sheet radargram analysis.

Hameed Moqadam, Daniel Steinhage, Adalbert Wilhelm, et al. Going deeper with deep learning: automatically tracing internal reflection horizons in ice sheets. ESS Open Archive . October 25, 2024. DOI: 10.22541/essoar.172987463.39597493/v1

How to cite: Moqadam, H., Bojesen, T. A., and Eisen, O.: Autoregressive mark-tracing for radiostratigraphy: A lightweight model for annotating internal reflection horizons in ice sheets, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17824, https://doi.org/10.5194/egusphere-egu25-17824, 2025.