Glacier Melt Simulator

A scenario model of the debris-covered ice cliff, ice cave and broken ice blocks in three field photos. Move the sliders or load your own weather and melt data, then scrub through the season. Two real photo reconstructions of the site are already loaded in the 3D view.

Cross-section through the cave

Reference photos

Until you load real before and after point clouds, the scene is built by eye from these three photos. Cliff height, cave size and block sizes are estimates you can change in Model settings.

What this model does and does not do
  • Surface melt uses an enhanced temperature-index scheme: melt = TF × air temperature + SRF × (1 − albedo) × sunlight, only above a temperature threshold (after Pellicciotti et al., 2005). Snow melts first, and fresh snow raises albedo until it is gone.
  • Debris cuts melt by a series-resistance rule, melt fraction = 1 / (1 + thickness / h*), which reproduces the usual fall-off of melt with debris thickness (the Östrem curve). The thin-debris boost is not included.
  • Ice cliff backwasting is the melt on the exposed face with a sunlight exposure factor. Extra melt at the base cuts an overhang; when it reaches a set fraction of the cliff height the wedge falls, the top edge steps back and new blocks appear.
  • Ice cave roof thins from above (debris-covered melt) and from below (warm air and stream water). It collapses when the roof is thinner than span divided by a span-to-thickness ratio.
  • Blocks shrink as cubes by the melt on their exposed faces.
  • Real before and after loads two VGGT point clouds that share one frame. Change colours points that have no neighbour in the other cloud within the chosen radius. The cross-section measures how far the front face moved inside a thin slab. Blend moves each point from the before surface to the nearest after surface as the model’s blend value rises. That is a nearest-surface match, not tracking of real objects, and points with nothing nearby dissolve instead. Lengths depend on the camera-height assumption of 1.5 m (about ±20%), and VGGT depth on snow and ice is a documented weak spot.
  • Weather defaults to monthly mean air temperatures for the Portage Glacier Visitor Center (NOAA 1991–2020 normals: June 10.6, July 12.2, August 11.1, September 8.3 °C), smoothed through the year, with synthetic day-to-day scatter, cloud and rain on top. Individual months in the synthetic year can run about 1 °C off the normals, as real years do; New weather draws another. It is one valley station, not your glacier. Use the warming shift to correct for elevation, or load your own series.
  • Before → after runs the model across your two photo dates and scans the melt scale k (a multiplier on TF and SRF) to find the range that gives a collapsed or standing cave roof and, if you enter it, the observed cliff retreat. A photo pair is a pass/fail constraint on k, not a measurement, so the result is a range. The blend curve says how far each part of the scene has moved from the first state to the second at each day.
  • Default values are typical orders of magnitude, not measurements from your glacier. Calibrate TF and SRF with the data loader before reading numbers literally.
  • Not modelled: ice flow and emergence, rain heat, sublimation, englacial and stream hydrology, real calving mechanics, and variation across the section. This is a scenario tool for exploring sensitivity, not a forecast, and not a safety tool: do not use it to judge whether an ice cliff or cave is safe to approach. Ice caves and cliffs collapse without warning.