A physical model of ice formation
413 routes across 4 venues, modelled twice a day from public weather data and shown beside reports from people who were actually there.
What it estimates, and what it never says
- Whether ice has formed
- A quantity of ice on a specific line, with a confidence value beside it. That is the whole claim.
- Never: quality, bonding, or whether to climb
- None of those are observable from weather data. A model that reported them would be inventing them. Ice can be present and lethal, and nothing here distinguishes the two.
Why not a degree-day model
Nearly every casual attempt at this counts freezing degree days. That is borrowed from lake ice, where a static sheet thickens by conducting heat upward and thickness follows the square root of accumulated cold. A frozen waterfall does not work that way.
Waterfall ice accretes. Water arrives near 0 °C and dumps its latent heat straight to the atmosphere, so growth is limited by how fast that heat can be removed — which depends on wind and clear-sky radiative cooling at least as much as on air temperature. A windy night grows more ice than a calm one at the same temperature, and a clear night more than an overcast one. The model reproduces both, and the test suite asserts them.
Why some routes say “withheld”
Aspect — which way a face points — is the largest term in the model. A route whose position is not confirmed would produce a confident wrong answer rather than an obvious gap, so those routes are listed with the reason and no estimate. Roughly two in five are withheld today, and each one says what would resolve it.
Reports come from people, not from the model
Three things stay separate on every route page and are never blended: cautions in this site’s own voice, quantities from the model, and observations attributed to whoever filed them. The model is bottlenecked on that last one — it cannot be fitted against anything else, which makes a report from a route you were on the most useful thing anyone can contribute.