Skip to main content
Site under development Estimates come from global weather models and have not yet been validated by a meteorologist. Do not use them as your only reference for safety decisions.
AlertaParques
ES

Methodology and evidence

AlertaParques is, for now, an informational and transparent weather tool that builds evidence over time.

AlertaParques estimates conditions for 82 trails in Chilean national parks using Open-Meteo data. This page explains how, which limits we know of and how we are gathering evidence to correct the thresholds. The exact rules and thresholds are in Criteria.

Models and confidence

  • The level is always decided by one model: Open-Meteo’s “best_match” option, which picks the model for each point automatically.
  • For the 15 volcano and altiplano trails, GFS (NOAA, gfs_seamless) and ICON (DWD, dwd_icon_global) are also queried. They only measure whether models agree (divergence); they do not change the level.
  • Confidence: low on altiplano above 3,500 m; medium on volcanoes above 2,000 m; high elsewhere. If the three models differ a lot, confidence drops.

Trails by terrain type: trail 55, mountain range 12, volcano 10, altiplano 5 (database, 2026-10-07).

The most important limit: the point’s elevation

Open-Meteo provides the estimate for the elevation of a 90 m terrain model at the queried coordinate, uses it to pick the model cell and adjusts values accordingly ( documentation ). In the mountains this produces two opposite errors:

  • Point lower than the exposed section: wind is underestimated. This is the pattern in the O Circuit cases.
  • Point near a volcano summit: the estimate describes the top, not the slope you are on, and in the cases reviewed wind was also far below what was reported on the ground.

That is why every park page shows the estimate point’s elevation next to the trail’s elevation.

Review of past incidents (backtesting)

To test the thresholds we queried historical data (ERA5, via Open-Meteo) at the place and approximate time of 10 incidents between 2023 and 2026: CONAF preventive closures and accidents, some fatal. We applied the rules and compared the level with what happened. We do not publish names or personal details of victims.

O Circuit — Torres del Paine, 17 Nov 2025 ■ Danger

A snow and wind storm between Los Perros and John Gardner Pass killed five hikers by hypothermia. ExplorersWeb reported 193 km/h gusts; other outlets reported up to 150 km/h. We know of no published official measurement. ERA5, at a point well below the pass, recorded a maximum gust of 108.7 km/h. With the O Circuit’s 80 km/h threshold, the system would have shown Danger from 05:00 for 16 hours.

Sources: ExplorersWeb, BioBioChile.

Llaima Volcano — Conguillío NP, 17 May 2026 ○ No alert — miss

A hiker died after falling about 600 m during the ascent, at around 2,000 m. Wind kept the helicopter away for hours. ERA5, at a point at 2,621 m, recorded a maximum gust of 41.8 km/h and −8.1 °C. With the criteria at the time, the system would have shown “no alert” all day.

Source: Meganoticias.

Villarrica Volcano — 30 May 2025 ○ No alert — miss

A hiker died after falling and rolling more than 400 m above 2,000 m. The helicopter could not land because of wind and bad weather. ERA5, at a point at 2,828 m, recorded a maximum gust of 43.6 km/h and −14.6 °C. With the criteria at the time, the system would have shown “no alert”; with the current volcano rules, −14.6 °C in May gives Danger.

Source: CNN Chile.

The other seven cases

Additional backtesting cases
Date Place System level What happened Source
21 Jul 2023 Circuito O ○ No alert Park closure due to snow and ice on routes. ERA5: 49 km/h gust (red threshold 80). Las Torres Patagonia, 26 jul 2023
3 Aug 2023 Lonquimay Volcano ▲ Caution Hiker died after a fall. ERA5: −9.5 °C, 12 km/h gust. Terrain hazard not captured by the model. BioBioChile, 3 ago 2023
19 Jun 2024 Villarrica Volcano ■ Danger Mountaineer missing; found dead in August. ERA5: −10.4 °C in winter → Danger by temperature. CNN Chile
1 Aug 2024 Villarrica Volcano ▲ Caution Preventive closure of Villarrica NP. ERA5: −5.4 °C, 54 km/h gust. The winter temperature rule left the level at Caution before gusts were evaluated. Diario La Quinta, 2 ago 2024
1 Aug 2024 Sierra Nevada (Conguillío) ■ Danger Preventive closure of Conguillío NP. ERA5: 80.3 km/h gust → Danger. Diario La Quinta, 2 ago 2024
23 Mar 2026 Circuito O ○ No alert O Circuit closure due to wind, snow and ice (CONAF announcement of 27 March). ERA5: 49 km/h gust (threshold 80). 24 Horas
21 May 2026 Osorno Volcano ■ Danger Hiker died after falling into a glacier crevasse. ERA5: 45.7 km/h gust → Danger. Radio Agricultura, 21 may 2026

In all these cases the real level was taken as Danger (official closure or victim). In October 2026 we removed two O Circuit cases (1 August 2024 and 1 August 2025) because we could not find a public source for them; the sample went from 12 to 10.

Weather model alone (ERA5) hits

40 % (4/10)

Hit = the system would have shown Danger when the real level was Danger.

Estimated coverage of the full system

~60–70 % (6–7/10)

An estimate, not a measurement: adds the cases the CONAF closures RSS would have shown. That RSS was integrated in July 2026, after most cases.

Ten cases are a small sample chosen after the fact: they help spot patterns (wind in Patagonia and on volcanoes is underestimated), not to claim an accuracy rate. This is a retrospective review; forward-looking verification is described below.

CONAF closures as a second source

Since July 2026, every hour the system reads the conaf.cl RSS, looks for trail-closure keywords, maps the region mentioned to its parks and shows the closure for 48 hours with a link to the notice. Limits: it does not detect closures announced only by radio, phone or signage; if the text does not mention the region it is not matched; and a regional closure may not affect every park in that region.

How we build evidence

Since 2 July 2026, every hourly estimate for every trail is stored in a verification log (alertas_historial table): issued level, each variable’s values and, for volcano and altiplano trails, the three models’ values. The aim is to later compare this log with what actually happened (official closures, ranger reports, incidents) and use that evidence to adjust thresholds by terrain type and seek professional validation.

There is not yet enough data to publish hit rates by terrain type.

Today the only automatic verifications come from CONAF closures matched by region. A regional closure is not always weather-related (e.g. fire risk) and does not affect every trail, so it cannot yet serve as “what actually happened” to measure hits. We also found hours with duplicated evaluations, which we will fix before computing rates.

Loading log status…

Were you on the ground?

If real conditions did not match the estimate — in either direction — tell us the park, trail, date, time and what you saw. It is the evidence we most need: alertaparques@gmail.com.

We are looking for meteorologists to collaborate

The current thresholds come from the incident review described here and have not been validated by a meteorologist. If you work in mountain meteorology, rescue or guiding and would like to review the criteria or how to verify them, write to us. When a validation exists, we will say so on this page and in the site notice.

Last update of this page or the rules: 7 October 2026 · See thresholds

How to use this information

Use it to prepare your trip and to know what to ask at the ranger station. “No model alert” does not mean it is safe, especially on volcanoes and in Patagonia. Always check CONAF and SERNAGEOMIN.

Weather data: Open-Meteo.com (CC BY 4.0), including ERA5 (Copernicus/ECMWF) for backtesting.