Calibrated probabilities
Probabilities are calibrated across successive historical periods, then checked on 2021–2024. Expected calibration error is 6.92%, so the number is designed to mean more than a simple snow ranking.
1 km local snow intelligence
Christmas snow odds for every corner of Britain.
A local probability for 43,686 towns, villages and postcode-linked locations, built from 34 British winters at 1 km resolution and sharpened with live global ensemble forecasts as Christmas approaches.
36.42%
lower probability error
Against the benchmark forecast across 151,980 out-of-sample results from Christmas 2021–2024.
151,980
Recent validation checks
Four Christmases assessed across 37,995 HadUK-linked locations.
0.961
Location discrimination
Exceptionally strong separation between Britain’s higher- and lower-snow locations.
17/17
Geographic checks passed
Positive skill in all 12 UK regions and all five spatial blocks.
Built for Britain
From mountain villages to city postcodes.
Britain's snow climate changes quickly over short distances. Elevation, latitude and exposure can separate two nearby places by far more than a national forecast suggests. The model preserves that local structure instead of flattening the country into broad regional averages.
1 km
source weather grid
Fine enough to preserve upland, valley and coastal snow patterns.
34
Christmas seasons
A consistent 1991–2024 historical foundation.
0.728
observed spatial correlation
Against the Met Office December snow-lying field.
3
global ensemble families
ICON, GFS and ECMWF guidance near Christmas.
Strong local discrimination
A 0.961 ROC–AUC shows that the model is exceptionally effective at distinguishing Britain's snow-prone locations from places where a white Christmas is genuinely rare.
Nationwide consistency
Validation remained positive in every standard UK region and every broad spatial block, from the Scottish Highlands to southern and eastern England.
The model
A 1 km portrait of Christmas snow.
Every local probability begins with the same nationwide weather record, interpreted consistently across Britain and resolved at a scale that keeps the geography visible.
Stage 01
Map 34 British winters
Daily temperature and rainfall from 1991–2024 create a consistent picture of Christmas weather across the entire country, using the Met Office HadUK-Grid record.
Stage 02
Identify snow, not just precipitation
Wet-bulb temperature and a published precipitation-phase relationship distinguish likely snowfall from rain, producing a snow signal for Christmas Eve and Christmas Day.
Stage 03
Keep the local pattern
Each 1 km cell retains its own history. A statistical prior steadies probabilities in places where snow is rare while preserving the differences created by latitude, elevation and exposure.
Stage 04
Turn the signal into a probability
Successive historical test periods refine the probability scale. The final percentage stays faithful to the physical geography while giving every location a clear, comparable chance.
Current forecast layer
Current weather joins the model as Christmas approaches.
Within 30 days, more than a hundred ensemble scenarios from ICON, GFS and ECMWF reveal the likely outcome and the uncertainty around it. Temperature is adjusted for local elevation, and the live influence rises as Christmas draws nearer.
ICON + GFS + ECMWFElevation-adjustedUpdated every six hours
30
day forecast window
The live layer begins as soon as Christmas enters ensemble range.
50%
maximum live influence
Even at close range, the 34-winter climate record remains part of the answer.
The live layer is deliberately capped at 50%, keeping the long historical record in the forecast while the shorter live-validation archive grows.
43,686
towns, villages and localities
37,995
local model cells represented
37,523
cells in the snow-lying check
430
independent GHCN station reports
Reading the forecast
A probability, not a promise.
The percentage is a calibrated local estimate, not a snow gauge at every doorstep. The historical layer reconstructs likely snowfall from daily temperature and rainfall, so an isolated shower or a very small mountain microclimate may not be captured.
The local pattern has been checked across 37,523 observed snow-lying cells, with a spatial correlation of 0.728, and sense-checked against 430 independent GHCN station reports. The full methodology and scorecard remain public.
Explore the model statisticsSee the forecast for your location.
Search for a town, village or postcode area, or compare probabilities across the UK map.