Small icebergs add a lot of Greenland fjord freshwater
High-resolution UNet mapping finds far more small icebergs than Sentinel-2 thresholding, raising Sermilik meltwater estimates toward ~2000 m³ s−1.
Source
Significant contribution of small icebergs to the freshwater budget in Greenland fjords
Study at a glance
- Design
- Other — UNet mapping of icebergs in Planet imagery vs Sentinel-2 thresholding in Greenland fjords
- N
- N=10000 · ~10,000 manual iceberg labels used to train the UNet
- Population
- Icebergs in Sermilik and Kangerlussuaq fjords, East Greenland
- Outcome
- Higher small-iceberg area and meltwater discharge estimates (~1300–2000 m³ s−1 Sermilik)
Structured fields used in claim comparison tables when every cited study has a complete layer.
Inspect source passages
Small icebergs add a lot of Greenland fjord freshwater
These passages occur verbatim in the stored paper text. They support the points listed below; they do not verify every statement in the explanation.
Authors manually annotated 10,000 iceberg instances.
We manually annotate a total of 10,000 instances of icebergs in those sub-images and perform elastic deformation and random rotation to augment the training data to a total of 1800 sub-images (over 40,000 iceberg instances, see Section S2.2).
How many more small icebergs appear in Planet imagery?
However, the total number of small icebergs in Planet imagery is nearly five times greater than those in Sentinel-2 (Fig.
What they did
Authors trained a UNet on ~10,000 manual iceberg labels in ~3 m Planet imagery of Sermilik and Kangerlussuaq fjords, compared with same-day Sentinel-2 TOA thresholding, fit size distributions, and converted area to meltwater with empirical keel/width relations.
What they found
P-UNet mapped 60.4 km² of icebergs in Sermilik, ~50% more than S-TOA, with ~5× more small bergs. Size distributions follow a power law only above ~300 m². Sermilik discharge is ~1310 m³ s−1 along-fjord and ~2020 m³ s−1 in a June–July comparison, exceeding some prior melt estimates; small bergs melt near the surface.
The limits
What it doesn't show
Two fjords and a few summer images do not give an annual cycle. Melt rates ignore fjord stratification and deep keels. Power-law extrapolation of small-berg contribution year-round is an assumption.
Key terms
- P-UNet
- UNet convolutional network trained on Planet imagery to segment icebergs.
- S-TOA
- Sentinel-2 near-infrared reflectance threshold used to pick icebergs.
- Power-law size distribution
- Fractal iceberg area-frequency relation that fails below ~300 m².
- Keel depth
- Submerged iceberg thickness estimated from surface length.
- Subglacial discharge
- Meltwater exiting at the glacier grounding line, compared here with iceberg melt.
Flashcards
0 of 11 answers reviewed
Research intelligence for this paper
See its role on concept claims, tensions it is part of, placement history, and related discoveries.
Quiz yourself
P-UNet Sermilik iceberg area is about:
Common questions
Why miss small icebergs in 10 m imagery?
They occupy few pixels and are too dark for simple reflectance thresholds.
Do small bergs change freshwater location?
They melt near the surface, unlike deep-keeled large bergs.
Is iceberg melt bigger than subglacial discharge?
On an annual scale in Sermilik, net iceberg freshwater exceeds subglacial discharge.
What size breaks the power law?
About 300 m².
More on Ice sheets