AI-ready Snow Radar Echogram Dataset (SRED) for climate change monitoring

TL;DR

This paper tackles the need for standardized radar echogram data to advance deep learning methods for internal snow-layer tracking in polar regions. It introduces the Snow Radar Echogram Deep Learning Dataset (SRED), derived from 2012 Operation Ice Bridge Snow Radar data, containing 13,717 labeled and 57,815 weakly-labeled echograms across diverse snow zones and resolutions, with a rigorous train/validation/test split. The authors formalize the layer-tracking problem, compare five segmentation approaches on binary and multi-layer formulations, and evaluate them using metrics directly tied to snow accumulation (N-pixel accuracy and MAE) as well as standard segmentation scores (ODS/OIS). Key findings show that while some models excel in dry zones, performance degrades in wet zones, highlighting the need for end-to-end architectures capable of directly estimating snow accumulation from echograms and motivating active learning to broaden geographic coverage. The SRED benchmark provides a valuable resource for fair comparisons, reproducibility, and accelerated progress in radar echogram analysis and climate monitoring, with potential extensions to Antarctica and broader climate applications.

Abstract

Tracking internal layers in radar echograms with high accuracy is essential for understanding ice sheet dynamics and quantifying the impact of accelerated ice discharge in Greenland and other polar regions due to contemporary global climate warming. Deep learning algorithms have become the leading approach for automating this task, but the absence of a standardized and well-annotated echogram dataset has hindered the ability to test and compare algorithms reliably, limiting the advancement of state-of-the-art methods for the radar echogram layer tracking problem. This study introduces the first comprehensive ``deep learning ready'' radar echogram dataset derived from Snow Radar airborne data collected during the National Aeronautics and Space Administration Operation Ice Bridge (OIB) mission in 2012. The dataset contains 13,717 labeled and 57,815 weakly-labeled echograms covering diverse snow zones (dry, ablation, wet) with varying along-track resolutions. To demonstrate its utility, we evaluated the performance of five deep learning models on the dataset. Our results show that while current computer vision segmentation algorithms can identify and track snow layer pixels in echogram images, advanced end-to-end models are needed to directly extract snow depth and annual accumulation from echograms, reducing or eliminating post-processing. The dataset and accompanying benchmarking framework provide a valuable resource for advancing radar echogram layer tracking and snow accumulation estimation, advancing our understanding of polar ice sheets response to climate warming.

Figures (6)

• Figure 1: Enhanced Snow Radar echogram collected in 2012, illustrating presumed annual accumulation stratigraphy
• Figure 2: Plot showing (a) original echogram with topography-elevation effects (b) surface flattened echogram (c) filtered echogram
• Figure 3: Map of Greenland showing the geolocation of flight lines used for dataset echograms and neighboring ice cores. Blue lines indicate the training data flight lines, while red, green, and yellow lines represent the L1, L2, and L3 test sites, respectively while black squares represent ice cores and shallow pits.
• Figure 4: (a) Echogram image (b) Binary segmentation map (c) Multi-class segmentation map
• Figure 5: Echogram image and model binary outputs (b) Ground truth annotation (c) UNet (d) AttentionUNet (e) DeepLab (f) FCN (g) Soft Ensemble