SyntheIce: A Synthetic Radargram Annotation Benchmark for Englacial Layer Reconstruction
Organizations: Department of Information Systems, University of Maryland Baltimore County (UMBC), Maryland, USA · iHARP, University of Maryland Baltimore County (UMBC), Maryland, USA
Abstract
Englacial layers in radio-echo-sounding radargrams (subsurface ice images) encode the climate and flow history of ice sheets and glaciers. However, large-scale extraction is hindered by time-consuming manual annotation and semi-automated detectors that produce fragmented rather than continuous horizons. We address two challenges: reconstructing complete layers from partial detections across radargram profiles and validating layer continuity across adjacent tile boundaries. We introduce a slope-aware greedy fragment-linking algorithm that reconstructs full-width layers using row-similarity gating and an exponentially weighted moving average (EWMA) slope predictor. Each reconstruction receives a content-based hash ID for stable cross-radargram identification. For quantitative evaluation, we generate SyntheIce, a synthetic dataset of 600 radargrams spanning multiple complexity levels and realistic noise conditions, including 150 wide radargrams divided into four adjacent tiles. All samples include exact ground-truth annotations. We evaluate the dataset using XCheck, a consistency-based validation framework, extending its original two-image setting to four consecutive tiles and enforcing continuity at every boundary. This demonstrates the dataset's utility for developing and evaluating real-world annotation algorithms. Our findings extend beyond ice science: when spatial features are fragmented but retain consistent local geometry, trajectory-based linking is more reliable than appearance-based matching. Evaluation against expert-labeled real radargrams demonstrates high precision, recall, F1-score, and mean IoU, supporting the method's transfer to real-world data.
Figures & tables
| Real Greenland Radar Datasets | SyntheIce Categories (This Work) | ||||||
| Statistic | Snow Radar (OIB/SRED) [ 27 ] | Depth Sounder (MacGregor) [ 1 ] | ICE2012 (Rahnemoonfar) [ 21 ] | Close-to Synthetic | Mid Level | Close-to Real | Noisy Level |
| Layers per radargram | 15–25 | 10+ | 12–25 | 15–19 | 18–21 | 23–26 | 35–47 |
| Fragments per layer | 1.4–1.8 | 1.0–1.5 | 1.6–2.0 | 1.5–3.0 | 2.0–4.0 | 3.0–5.0 | 3.5–5.0 |
| Fragment length(px) | 200–600 | 300–800 | 250–500 | 1001–1769 | 765–1418 | 541–1069 | 442–844 |
| Annotation width(px) | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| Terrain tilt ( ∘ ) | 0–12 | 0–8 | 0–10 | 1.0–2.8 | 1.1–2.9 | 1.2–3.0 | 1.5–3.0 |
| IoU 0.20 | IoU 0.30 | IoU 0.50 | IoU 0.70 | |||||||||||||
| SyntheIce Category | P | R | F1 | mIoU | P | R | F1 | mIoU | P | R | F1 | mIoU | P | R | F1 | mIoU |
| Close-to-Synthetic | 0.96 | 1.00 | 0.98 | 0.99 | 0.96 | 1.00 | 0.98 | 0.99 | 0.96 | 1.00 | 0.98 | 0.99 | 0.95 | 0.98 | 0.96 | 1.00 |
| Mid-Level | 0.95 | 1.00 | 0.97 | 0.98 | 0.95 | 1.00 | 0.97 | 0.98 | 0.95 | 1.00 | 0.97 | 0.99 | 0.93 | 0.97 | 0.95 | 0.99 |
| Close-to-Real | 0.88 | 1.00 | 0.93 | 0.91 | 0.88 | 1.00 | 0.93 | 0.91 | 0.84 | 0.95 | 0.89 | 0.93 | 0.77 | 0.87 | 0.81 | 0.96 |
| Noisy-Level | 0.91 | 0.99 | 0.95 | 0.85 | 0.90 | 0.98 | 0.94 | 0.86 | 0.82 | 0.89 | 0.85 | 0.90 | 0.70 | 0.76 | 0.73 | 0.96 |
| Precision | Recall | F1 | Mean IoU | |
| 0.20 | 0.828 | 0.995 | 0.900 | 0.808 |
| 0.30 | 0.817 | 0.985 | 0.889 | 0.809 |
| 0.50 | 0.776 | 0.929 | 0.842 | 0.821 |
| 0.70 | 0.730 | 0.884 | 0.795 | 0.838 |
Remote Sensing Benchmarks
Remote sensing benchmarks measure how well models interpret Earth-observation data across sensors, regions, time, tasks, and operating conditions.
See all 218 papers