Glacier ObservatoryGEOSPATIAL MODEL EXPLORER

HINDU KUSH HIMALAYA

Every prediction has a landscape.

Explore glacier segmentation, find failure modes, and compare the evidence.

Research collectionPaper · 2022 · Restored demo · 2026

01 / GEOGRAPHIC CONTEXT

Explore the landscape

PATCH ACCURACY
50%75%100%
Loading geographic data…
147 patches in viewClick a marker to inspect a patch
SYNCED ZOOM1×
A
Satellite imagePreprocessed model input
INPUT
Satellite input for selected glacier patch
B
Reference labelGlacier inventory annotation
REFERENCE
Reference segmentation for selected glacier patch
C
Model predictionU-Net segmentation result
PREDICTION
Predicted segmentation for selected glacier patch
Clean iceDebris-coveredBackground
Zoom together. Drag any image to pan.

RESEARCH, PRESERVED

About this collection

This explorer brings together the publicly displayed results from Interactive Visualization and Representation Analysis Applied to Glacier Segmentation, by Minxing Zheng, Xinran Miao, and Kris Sankaran.

147 patchesInput, reference, prediction, and displayed accuracy recovered from the original interactive app.
A clearer visual languageThe Glacier palette recolors the saved masks while preserving class assignments. Cyan: clean ice. Sand: debris-covered. Choose Original paper to see the archived colors.
Geographic contextPatch coordinates are read from the original app’s map popups. Glacier outlines come from the project’s archived vector data.
Visual difference layerComputed from the displayed masks for inspection. It does not replace the original model’s pixel accuracy or provide new IoU estimates.

This is a 2026 reconstruction of the original interactive results. Reported accuracy is copied at the precision displayed by the original app. Patch imagery is preprocessed model input and may differ from natural-color satellite basemaps. The map is geographic context; patch images are not georeferenced overlays.

Read the paper Original code Original panel