Cairngorm Summit · 1,245m · Heriot-Watt University AWS · Since 1990
The Cairngorm
Observatory
Thirty-five years of half-hourly readings from Britain’s highest automatic weather station — temperature, wind and snowpack data from one of the most extreme environments in the British Isles, turned into stories you can see and explore.
Temperature · 1990–present
When Does Winter Come?
We define “winter” as the first sustained period where the 14-day rolling mean temperature stays below 0°C. Each bar is one season, running from its autumn onset to the following spring thaw. Greyed rows indicate years with data gaps.
The summit station has long gaps — no usable data for 1995, 2001–03, 2010, 2014 or 2019–24. Where it went quiet we estimate the daily temperature from Met Office HadUK-Grid, calibrated against the station itself. That takes the record from nineteen complete seasons to thirty-five. Hatched bars are estimates, not measurements, and the toggle above will drop back to station-only data at any time.
Gaps in the station record filled from Met Office HadUK-Grid, calibrated to the summit. Hatched bars are estimates.
Wind · Summit Records
The Wind Rose of the Plateau
Cairngorm’s summit is one of the windiest places in Britain. Each petal shows the frequency of wind from that direction — inner bands are calmer winds, outer bands are gales. The dominant SW/W airstream and the devastating northerlies are both visible.
Extreme Wind Events
The Gust Record
Every gust over 100 mph recorded at the summit, plotted against named storms and the calendar. Readings above the recognised UK record of 173 mph — itself set at this station in 1986 — are excluded rather than presented as a record we cannot claim.
About
Data & Methodology
About the Data
Weather data: Cairngorm Automatic Weather Station, operated by the Heriot-Watt University Physics Department at 1,245m on the summit of Cairn Gorm. Half-hourly readings of wind speed, wind direction and two independent temperature sensors. Freely available at cairngormweather.eps.hw.ac.uk. Please reference Heriot-Watt University as the data source.
Data quality: Only readings where both sensors report healthy status are used. Temperatures are flagged as suspect when mean wind drops below 10 mph, because the sensor housing aspirates poorly in still air. Monthly means are suppressed where fewer than 96 valid readings exist, so a handful of readings never masquerades as a month.
Wind units: the archive records metres per second for 1990–1995 and again from 2016 onward, but miles per hour in between. All figures here are converted to mph.
Wind rose: all valid wind direction readings aggregated into 16 sectors of 22.5°, banded 0–20, 20–40, 40–60 and 60+ mph.
Filling the gaps
The summit station has been offline for whole seasons — 1995, 2001–03, 2010, 2014 and 2019–24 — plus scattered missing days. Where it has no usable reading we estimate the daily temperature from Met Office HadUK-Grid, a 1 km gridded dataset built by interpolating the UK weather station network. Wherever the summit station did record, its reading is always used; the grid only ever fills blanks.
A 1 km grid square cannot capture an exposed 1,245 m summit, so we calibrate it against the station month by month, using the 5,859 days where both recorded. Raw, the grid runs 0.7–1.3 °C colder than the summit in every month of the year. After calibration that bias falls to within ±0.15 °C, with a daily RMSE of 1.62 °C and a correlation of 0.946.
How well does it recover a season?
Getting the daily temperature right is not the same as getting the season right. To test that, we hide each complete season in turn, refit the calibration without it, and see whether the model recovers that season’s onset and thaw.
Across all eighteen testable seasons the average error is 8 days for the onset and 24 days for the thaw. But that understates the onset and overstates both, because it scores the model against station dates that are themselves sometimes built from thin data. Restricted to the nine seasons where the station recorded at least 80% of days, the onset error falls to 1.1 days and the thaw to 13.3 days.
Treat modelled thaw dates with more caution than onsets
The onset is a sharp transition — autumn cools quickly, so the model lands within a day or two. Spring is different. The summit hovers around freezing for weeks, thawing and refreezing, so the day the 14-day mean finally crosses zero and stays there is very sensitive to small errors. A 1.6 °C difference that barely matters in November can move a thaw date by a fortnight. Lying snow makes it harder still: a snow-covered summit stays colder than the surrounding landscape the grid is interpolated from, and HadUK-Grid measures air temperature, not snow.
Modelled onsets are good estimates. Modelled thaws are indicative — the middle of the distribution is accurate to a couple of days, but the tails are long.
Season basis. A season is marked observed when under 5% of its days are filled, part modelled between 5% and 50%, and modelled above 50%. Hover any bar to see which it is and how much was filled. Where the station recorded dates of its own, the tooltip shows those too, so you can compare.
Trends. Trend lines are only drawn where the fit is statistically significant, and the thaw is held to a stricter bar than the onset for the reasons above. Both the station-only and gap-filled series are available through the toggle, so any claim can be checked against observations alone.
HadUK-Grid data: Hollis, D., Carlisle, E., Kendon, M., Packman, S., Doherty, A. (2026). HadUK-Grid Gridded Climate Observations on a 1km grid over the UK, v1.3.2.ceda (1836–2025). NERC EDS Centre for Environmental Data Analysis. doi:10.5285/789b3065d74a4c948ab05d33556c86d0. Contains public sector information licensed under the Open Government Licence v3.0.
Visualisation created for Cairngorm Ski Club. Run observatory/data/process_aws.py to regenerate the dataset from the latest archive.