For informational purposes only. Do not rely on it for your safety.

How the model works

A rebuild of a 2016–2018 Excel model for the Adirondack High Peaks, where no avalanche center issues a forecast.

In one paragraph

The app pulls hourly weather for a point at a chosen elevation (temperature, humidity, wind, precipitation, cloud, radiation) from public archives and forecasts, decides each hour whether precipitation fell as snow or rain, converts snow water to depth with a temperature-keyed snow-to-liquid ratio, and stacks the result into layers. Every day it settles the layers, melts and refreezes them, checks the bulk temperature gradient for faceting, grows surface hoar on clear calm cold nights, drifts loose snow into wind slabs on lee aspects, and records the processes it believes ran. From the last 24–72 hours and the next 24–48 it evaluates the classic forecaster red flags, groups them into avalanche problems (type, likelihood, size, aspects), and reads a danger level off a hazard chart. Every number here is modelled; none is observed.

Lineage

The original workbook refreshed five web queries (NWS observations at Saranac Lake, the NWS hourly digital forecast for Mt. Marcy, mountain-forecast.com, NOHRSC snow depth, sunrise/sunset), derived per-hour snow quantities in two “Formatted” sheets, aggregated them into a “Forecast Model” indicator board for the last and next 24 hours, and archived one summary row a day into a “Snowpack Profile” via a macro. Its heuristics are preserved here: 12 in / 1 in per hour / 1 in SWE / 6 in of dense snow for storm slabs, 15–20 mph for wind transport with lee aspects opposite the dominant wind, a 15 °F rise in 12 hours for rapid warming, a 10 °C/m gradient for faceting, clear-calm-cold nights for surface hoar, and rain or three days above freezing for a snowpack reset. What changed: weather now comes from elevation-downscaled gridded models rather than a valley airport; the snowpack is a persistent layer stack rather than a daily journal; a few spreadsheet bugs were fixed (a reversed sign that flagged rapid cooling as rapid warming in the forecast sheet, a broken density lookup, rain counted as snow water); and indicators are combined into problems and a danger level.

Data

Reading the danger level

Problems are placed on a likelihood × size chart after the Conceptual Model of Avalanche Hazard (Statham et al. 2018) and the highest cell sets the level on the North American Public Avalanche Danger Scale. Weather-only rules catch most real events but also raise many false alarms, so treat the output as a list of questions to ask the snow, not as permission. Dig, look for recent avalanches, listen for collapses, and keep the slides for days when the model and the snow agree.

Replay

Any day of a stored season can be replayed from the date picker on a site page, and each replay has its own shareable address such as /snowpack/sites/adk-high-peaks/2022-02-12. The model rebuilds that season up to that morning and uses what actually happened next as the forecast, which means a replay shows more than anyone could have known at the time. It reads only weather already stored, writes nothing, and never calls a weather service, so it costs about a tenth of a second and the same date always gives the same answer.

Ten seasons are loaded, 2016-17 through 2025-26. Seasons from 2018-19 onward come from the archived 3 km model; 2016-17 and part of 2017-18 predate it and are reconstructed from a roughly 25 km reanalysis, which smooths the terrain away almost entirely. Those seasons are marked on the page and should be read as much weaker evidence. Weather back to 1940 exists in the reanalysis if an older winter is ever wanted.

The week around 12 February 2022 is the one to look at: two skiers remotely triggered a persistent slab on the Angel Slides of Wright Peak and one was fully buried and recovered alive. Replaying it is the closest thing this model has to a test against a known outcome, and it should be read with care. The model never saw that snowpack, and a single agreeing case picked after the fact is not validation.

Weather data by Open-Meteo (CC BY 4.0) and NOAA/NWS. Back to the forecast.