Remote Sensing, Glacier Calving and Iceberg Tracking

Think of this project like visualizing a moving coastline or tracking the path of a storm—problems that involve looking carefully at how something changes shape or position over time. In those familiar scenarios, you'd typically compare two snapshots: the coastline position today versus last year, or the storm's current track compared to its forecasted path. Similarly, in our glacier calving visualization, the "snapshots" aren't two dates, but rather two distinct versions of the same glacier front at a single moment: one is the real, observed glacier front (what actually happened), and the other is the front predicted by the automated method (the computer's guess). Abstractly, the task is the same—clearly showing differences between two shapes—but instead of tracking temporal change, we're highlighting differences in prediction accuracy. We want to focus on effectively visualizing how closely (or not) predictions match reality.

Dramatic video to set the stage =)

Intro to Satellite Imaging

What is Remote Sensing?

Iceberg Naming Info

• http://ice.aari.aq/docs/NIC_Iceberg_naming.pdf

Some Inspiration and Further Reading

https://tc.copernicus.org/articles/17/4957/2023/

https://tc.copernicus.org/articles/18/3315/2024/

Effect of bedrock topography on calving front variations for (a) the Ingia Isbræ, (b) the Kangerdlugssuaq Glacier, and (c) the Daugaard-Jensen Glacier. In each panel (from left to right), there is a satellite image showing calving front trajectories, a marked profile indicating bedrock topography, color-coded calving front positions along the profile, and the corresponding time series of calving front variations. Note that the axes are scaled differently in each panel. Landsat-8 imagery is courtesy of the U.S. Geological Survey.

Calving front of the Bowdoin Glacier: orthoimages were obtained by UAV on (a) 11 and (b) 16 July, with the resulting velocity field inferred using feature tracking (c) and the maximum principal directions of the strain rate (d), respectively. For the sake of visualization, only the maximum directions with a magnitude above 5 yr-1 are drawn on (d). The vector lengths scale with the magnitude of the maximal principal strain rate. A scale corresponding to 100 yr-1 is shown as reference. Blue and red arrows stand for principal directions on the upstream and downstream edge of the crack, respectively.

source: https://www.researchgate.net/figure/Calving-front-of-the-Bowdoin-Glacier-orthoimages-were-obtained-by-UAV-on-a11-and-b16_fig2_316935718

Visualization of the mean distance error (MDE) calculation between front label Q and front prediction P. The dist(q,P) represents the distance between q and the pixel in P that is closest to q.

https://www.researchgate.net/figure/Visualization-of-the-mean-distance-error-MDE-calculation-between-front-label-Q-and_fig8_375899741

How are the metrics computed? Read the paper

Where to get satellite data

• https://www.sentinel-hub.com/explore/eobrowser/
• https://browser.dataspace.copernicus.eu
• Planet Labs