🔒The experiment is closed – thank you!
Data collection has ended: 40 analysts picked the glacier surface and bed on the 15 radargrams. Many thanks to everyone who took part. The results, the anonymised picks and the radargrams are published here:
- PaperSiebenbrunner and others: The human factor in GPR-derived ice thickness: Inter-analyst variability as a benchmark for automated boundary layer picking – preprintcoming soon
- DatasetAnalyst picks, model picks, human consensus and radargrams – Zenodocoming soon
- CodeThis picking tool – github.com/anna7br/gpr-picking-experiment (MIT licence)
The picking tool stays online as a demo.
Open the picking tool →Load radargram images with Open images…: your own PNG/JPG files, or the 15 experiment radargrams from radargrams_png.zip in the Zenodo dataset. Images are opened locally and never leave your browser; picks are saved with Download my results (CSV) in VIA-compatible format.
Original participant instructions (archived)
ℹ️What this is
Thank you for taking part! We are measuring how consistently experts pick reflections in glacier GPR radargrams. On each image you draw two lines: the glacier surface and the bed. There are no graded "right" answers – we want your natural interpretation, so please pick as you normally would.
✏️How to pick
- Choose the horizon with the Surface (s) / Bed (b) buttons (or the s / b keys). Surface draws in cyan, bed in orange.
- Surface: left-click along the top reflection of the ice body.
- Bed: left-click along the ice–bedrock reflection – the bottom of the ice. Where the ice lies above an air-filled cavity, follow the bottom of the ice (the ice–air interface inside the cavity), not the rock below.
- Add points left to right. To move a point, drag it to a new position (grab it with the left mouse button); the line always reconnects left to right. Zoom with the mouse wheel, pan by dragging with the right mouse button. "Undo point" (Ctrl/Cmd+Z) and "Clear horizon" act on the horizon you are currently picking.
- If you cannot interpret the bed on an image, tick "bed not interpretable" (you can still pick the surface) and move on – please do not guess.
- If you are only sure in parts of the radargram, you can use a "straight" connector between areas of high confidence. During the analysis, such sections will be discarded automatically.
- Move between images with Next / Prev (or the arrow keys); your lines are kept. The counter in the top-left corner shows which radargram you are on (e.g. "Radargram 3 / 15") so you can see how many remain.
✅When you are finished
- When you have picked the last radargram, a "thank you" message appears in the tool. Click Download my results (CSV) – this saves a file named
analyst_YOURNAME.csv. - Upload that file using the green Upload results button (at the top of this page, or inside the picking tool). You will be asked to sign in to Google so the file is saved to our project drive. (The upload form is closed.)
🔎Please
- Work independently – do not discuss or compare picks with other participants.
- Pick both horizons on every image you can.
- Try to finish in one sitting: your work is held in the browser tab and is saved only when you click Download. Nothing is sent anywhere until you upload.
The tool runs entirely in your browser. You can return to these instructions anytime via the "? Instructions" button in the tool.
🙏Thank you
Your time and expertise are greatly appreciated – these picks directly support the evaluation of UAV-borne GPR ice-thickness measurements in the MELT.AI project. If you have any questions about the task or run into technical problems, please get in touch:
Anna Siebenbrunner
GEORESEARCH & Technical University of Munich
anna.siebenbrunner@georesearch.ac.at