Independent UK snow forecast

Historical outlook · Christmas 2026

Will it snow where you are?

Discover the odds for 43,686 UK towns and villages, powered by our 1 km snow model and 34 years of British winter.

Find your local forecast

116Days
00Hrs
37Min

National outlook

The Christmas snow map

11%

Chance of a widely white Christmas

Hover or tap a region
LowerHigher chance
Odds of a widely white Christmas by region · 4% at any single spotOpen detailed snow map →
The forecast desk

National ranking

Snowiest places this Christmas

Around the country

Explore by region

Method summary

What's behind the number

A 1 km observational climatology, a wet-bulb phase diagnostic, three global forecast ensembles and a North Atlantic Oscillation layer, calibrated together and scored on held-out Christmases.

1.29M

cell-Christmases analysed

37,995

1 km grid cells

122

live ensemble members

34

winters of record

Two different questions. Town, village and postcode pages answer “does it snow here?”. Area pages can't use that number — an average sits below every place inside it — and “does it snow somewhere?” is no better, since that rises the more finely you grid the country. So national and regional headlines give the odds of a widely white Christmas: falling snow reaching at least a tenth of the area's towns. Both figures are built from the same calibrated probabilities, so they stay on one scale.

01

HadUK-Grid 1 km observations

Met Office daily temperature and rainfall, 1991 to 2024, on a 1 km grid: 37,995 cells across 34 winters, or 1,291,830 cell-Christmases. Each of the 43,686 towns, villages and postcode districts resolves to its own cell, so elevation differences over a few kilometres are retained.

02

Precipitation-phase diagnosis

Rain and snow are separated by wet-bulb temperature, computed with the Stull approximation and combined with a published precipitation-phase relationship. A day counts as white if snow fell on 24 or 25 December.

03

Rolling-origin calibration

The probability scale is corrected against three rolling five-Christmas windows. The calibration is monotonic in log-odds, so the 1 km spatial ordering is preserved. Each place also carries a beta-binomial posterior interval.

04

Multi-model ensemble

Within 30 days, about 122 members from ICON, GFS and ECMWF IFS are pooled. Each model's native full-column snowfall field is blended 50/50 with our surface diagnostic, then combined with the climatology in log-odds space on a lead-time ramp capped at 0.5.

05

Sub-seasonal NAO layer

Between 12 and 27 days out, the national baseline is shifted by the North Atlantic Oscillation, read as a seven-day mean of the NOAA daily index. The coefficient is held at -0.25 log-odds per NAO unit; fitting it freely across 38,000 spatially correlated cells overstates the sample, which is 34 Christmases.

06

Forward-chained validation

Each Christmas is scored with the model rebuilt from prior years only. Brier score and log loss are reported together, over windows including 2009 and 2010, since a window without snow in it favours any forecast that lowers its probabilities.

Current scorecard. Over nineteen forward-chained Christmases from 2006, the daily-index NAO layer reduced log loss by 6.1% against no NAO layer, against 2.0% for the monthly index it replaced. Promotion required beating both on Brier and log loss across two test windows. See the full scorecard.

Forecasting is never certain, and long-range figures are guidance rather than a guarantee. For entertainment and general interest. This is not an official Met Office forecast.

Looking for something specific? Browse all UK regions →