Methodology
This page explains where the numbers on the map come from, what they mean and where they can mislead. Everything is computed by the open-source pipeline in this repository, so anyone can reproduce it.
The data: ERA5-Land
We use ERA5-Land, a global reanalysis of the land surface produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) for the Copernicus Climate Change Service (C3S). A reanalysis combines a weather model with millions of past observations to reconstruct the weather everywhere on Earth, hour by hour, back to 1950.
- Grid: 0.1° × 0.1°, about 9 km by 11 km in mid-latitudes (cells get narrower towards the poles). The map shows one colour per grid cell.
- Years: every December from 1950 to 2025 (76 years). A new year is added each January once the December data is published.
- Variable: snow depth, in centimetres of snow (not its water equivalent). ERA5-Land's snow depth is an average over the whole 9 km cell: when only part of the cell is snow-covered, the published depth is the snow spread over the whole cell. "5 cm" therefore means "5 cm of snow on average across the cell", which is not the same as a ruler reading at one spot.
- Daily values: we download the official daily statistics product
(
derived-era5-land-daily-statistics), which averages the 24 hourly values of each day. A day is a UTC calendar day (see limitations). - Days: 24, 25 and 26 December.
- Land only: ERA5-Land also simulates snow on lakes and coastal water. Cells that are mostly sea or lake (land-sea mask below 0.5, or lake cover of 0.5 or more) are left blank, so Lake Michigan or Lake Ladoga show as water rather than as land with little snow.
Snow depth is stored rounded down to whole centimetres, and thresholds are whole centimetres, so "at least 5 cm" means exactly that.
What counts as a white Christmas
A grid cell has a white Christmas in a given year when ERA5-Land's daily-mean snow depth reaches a threshold. The site offers two thresholds and three rules.
The threshold: 5 cm, calibrated to weather stations (default), or the raw 1 cm
Most people mean "snow on the ground", and weather services count that as at least 1 cm on a snow ruler. Taken literally, ERA5-Land at ≥ 1 cm is far too snowy: compared with weather stations it gives a white Christmas Day about 24 percentage points too often (Munich 72 % against about 34 % measured, Chicago 72 % against 39 %; see the station check below). ERA5-Land carries too much thin, patchy snow in the lowlands, a bias also reported in the scientific literature.
So the default threshold is 5 cm in ERA5-Land, the whole-centimetre value that best matches "at least 1 cm measured at a weather station". You can switch to the raw 1 cm view, which is ERA5-Land exactly as published. We show both because the calibrated view is the better estimate of what you would have seen outside, while the raw view is the unadjusted reanalysis that other studies and datasets use, and the difference between the two is itself a measure of the uncertainty. Future projections use the calibrated threshold only.
The three rules
| Rule | Condition | Notes |
|---|---|---|
| Snow on Christmas Day (default) | on 25 December | "Snow on the ground on Christmas Day." |
| Snow on all three days | on each of 24, 25 and 26 December | The German Weather Service (DWD) rule; the strictest. |
| Snow on any of the three days | on at least one of 24, 25 or 26 December | The most lenient. |
The probability of a white Christmas is the share of years in a period that were white. For example, 30 % over 1996–2025 means 9 of those 30 Christmases were white.
Checked against weather stations
We compare ERA5-Land with weather stations that measured snow depth on 25 December for decades,
from NOAA's Global Historical Climatology Network (GHCN-Daily). The check is part of the pipeline
(whitexmas stations), with the station list in config/pipeline.yaml, and it writes a full
report (snow_variable_check.html).
Observed: a station has a white Christmas Day when its snow depth on 25 December is at least 1 cm. US stations report whole inches, so for them it is "at least 1 inch as rounded by the observer", about 1.3 cm.
ERA5-Land: the same grid cell the map uses for that location, daily-mean snow depth on 25 December, over exactly the years the station covers.
Stations: 46 with at least 15 overlapping years. 30 are used for the calibration. Left out: 9 that are white almost every year (every threshold agrees there), 4 with patchy records (often reporting snow only when there is some, which inflates their frequency), and 3 whose cell is more than 200 m higher or lower than the station (valley towns under a mountain cell, such as Sion).
Result:
ERA5-Land threshold Mean difference to stations Typical error per station ≥ 1 cm (raw) +24.3 percentage points 24.3 pp ≥ 4 cm +4.6 pp 9.0 pp ≥ 5 cm (default) +0.3 pp 8.1 pp ≥ 6 cm −2.4 pp 8.4 pp At 5 cm, ERA5-Land and the station agree on white or green in 84 % of station-years. Repeating the choice 30 times, each time leaving one station out, picks 5 cm 29 times and 6 cm once.
Timing is not the cause: using the morning value instead of the daily mean (06 UTC in Europe, 12–13 UTC in North America) changes the difference by less than one percentage point.
Scope: the 30 stations are in Europe and North America, between 40° and 70° N, from sea level to 1,611 m (Denver). The calibration is not tested for high mountains, other climates or the Southern Hemisphere, where 25 December is midsummer. Individual places can still be off by about 8 percentage points either way, more in the semi-arid interior of North America, where ERA5-Land stays too snowy even at 5 cm.
Periods
- 1950–2025: the full record.
- 1961–1990 and 1996–2025: two 30-year "climate normals", for a then versus now comparison.
- Decades ending in the latest year: 2016–2025, 2006–2015, …, 1956–1965.
A probability is shown only where at least 80 % of the years in the period have data, which in practice means every land cell.
Change and trend
- Change is the difference between the two normals, 1996–2025 minus 1961–1990, in percentage points (pp). From 40 % to 25 % is a change of −15 pp.
- Trend is a straight line fitted (ordinary least squares) through the 76 yearly white/green outcomes, expressed in percentage points per decade.
- Is the trend real? Whether a Christmas is white depends heavily on the weather of a few days, so the yearly series is very noisy. We test each cell with the Mann-Kendall test, a standard test for a monotonic trend that makes no assumption about the distribution (with the correction for tied values, which matters for a series of zeros and ones). Where the test gives p ≥ 0.1, the trend could easily be chance, and the trend map hatches the cell. Cities in the search list show a trend only when p < 0.1.
- We do not show trends for single decades: a slope through ten yes/no values is mostly noise. For decades we show the probability itself, so you can compare them.
Glaciers and ice sheets
On glaciers and ice sheets (Greenland, Antarctica, Svalbard, high mountains) ERA5-Land carries a permanent, very deep snowpack (a fixed 10 m of water equivalent), so every Christmas is "white". Shown as 100 %, these areas would dominate the map and say nothing about Christmas weather. A cell is treated as glacier when both:
- ERA5-Land's own glacier mask covers at least half of the cell, and
- it had at least 1 cm of snow on at least one of the three days in every year from 1950 to 2025 (the same rule for both thresholds, so the glacier areas do not change when you switch).
Glacier cells are drawn hatched grey on every map layer and flagged in the place panel, and they are left out of the country averages.
Countries
Country figures average all land cells inside each country (Natural Earth borders at 1:10 million), weighting each cell by its area (cells shrink towards the poles, by the cosine of latitude), with glacier cells excluded. Countries are identified by their ISO 3166 three-letter code. Where Natural Earth draws a country as several pieces with the same code (for example Kazakhstan and the leased Baikonur area, or Australia and its small island territories), the pieces are counted together.
Places
The search list uses all places with at least 15,000 inhabitants from GeoNames. Each place takes the value of the grid cell it lies in. A coastal place whose cell is classed as sea or lake uses the nearest land cell within two cells (about 20 km); places with no land cell that close are left out.
Limitations: read this before quoting a number
- A 9 km cell is not your garden. The value describes the average over the whole cell. Your street can differ, especially near coasts, lakes and in cities, which are warmer than their surroundings.
- Mountains are smoothed. Each cell has one average height. A deep valley and the peaks around it share a cell, so valley towns can look snowier than they are, and summits less snowy. In the station check, Sion in the Rhône valley sits under a cell almost 1,000 m higher, and ERA5-Land calls every Christmas there white.
- The calibration is an average. 5 cm matches the stations on average; at a single place the difference is typically about 8 percentage points either way (see the station check).
- The day is a UTC day. Daily means run from midnight to midnight UTC. In Europe that is within an hour or two of local time, but in North America the "25 December" average runs from the evening of the 24th to the evening of the 25th local time, and in East Asia from the morning of the 25th to the morning of the 26th. The station check found this changes the numbers by less than one percentage point.
- A reanalysis is a model. ERA5-Land's snow is not measured directly: it is simulated from ERA5 weather, which assimilates observations. It tends to be less reliable where few weather stations exist.
- Earlier years are less certain. Fewer observations went into the reanalysis in the 1950s and 1960s, especially outside Europe and North America.
- The old maps are not comparable. An earlier version of this analysis counted a single hour with 1 cm of snow at any time over the three days, and counted the sea as 0 %. Its numbers are higher and should not be compared with these.
Licence and citation
- Data on this site: CC BY 4.0. Code: MIT. Source code and data downloads: https://github.com/Kochal/white_xmas.
- Contains modified Copernicus Climate Change Service information (ERA5-Land). Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains.
- ERA5-Land: Muñoz Sabater, J. (2019): ERA5-Land hourly data from 1950 to present. Copernicus Climate Change Service (C3S) Climate Data Store (CDS). DOI: 10.24381/cds.e2161bac. Muñoz-Sabater, J. et al. (2021): ERA5-Land: a state-of-the-art global reanalysis dataset for land applications. Earth System Science Data 13, 4349–4383. DOI: 10.5194/essd-13-4349-2021.
- Snow-depth bias of ERA5-Land in the literature: Monteiro, D. and Morin, S. (2023), The Cryosphere 17, 3617–3660, DOI: 10.5194/tc-17-3617-2023; Stachura, G. and Ustrnul, Z. (2026), The Cryosphere 20, 2295–2315, DOI: 10.5194/tc-20-2295-2026.
- Weather stations: Menne, M. J. et al., Global Historical Climatology Network – Daily (GHCN-Daily), NOAA National Centers for Environmental Information.
- Places: GeoNames (CC BY 4.0). Borders: Natural Earth (public domain).
To cite this dataset: White Christmas: probability of snow at Christmas from ERA5-Land, 1950–2025, data version as shown on the site. https://github.com/Kochal/white_xmas.