The Epiphany frost

After Epiphany the frost came, until about 1987

At all 19 long-record weather stations, the week of 6–12 January averaged 2.5 °C colder than the week before in 1953–1987. The gap then faded and from 1988 it is gone. Pick a town and see the winters of its nearest station.

Daily mean temperature (max + min) / 2, NOAA GHCN-Daily. Differences are measured against the station’s seasonal normal. The pattern is described in measurements; its cause is not established.

București-Băneasa: in 1953–1987, the week after 6 January averaged 1.9 °C colder than the week before, relative to the station’s normal; in 1988–2025 no cold step shows (+0.4 °C).

Every winter a row

Daily mean temperature minus the station’s normal, 20 December to 25 January, winter 1953 at the top, 2026 at the bottom; by default a moving average over 11 winters and 3 days, so that harsh or mild winters do not drown the rest. Blue: colder than normal; ochre: warmer. The vertical line is 6 January, the dotted one separates 1987 from 1988. Touch or hover over a row to see the winter, the day and the departure.

#A cold week, not a day

Each line is the daily mean departure from the station’s normal. In 1953–1987 the curve sits slightly above normal before 6 January (30 December–5 January average: +0.4 °C), then drops and stays below normal for the whole week of 6–12 January (−2.1 °C). The gap between the two weeks is the “step”: −2.5 °C on average, with about ±0.7 °C of uncertainty for a station’s mean. About a sixth of the step comes from the mild lead-in to January, the rest from the cold week.

In 1988–2025 the curve no longer drops: +0.5 °C before 6 January, +1.2 °C after. The mean step is +0.7 °C and at every station it stays within what 37–38 winters of chance can produce (no station exceeds twice its uncertainty). The conclusion stops there: the cold step is no longer visible.

#Winter by winter, 1953–2026

Each bar is one year: how much colder (down, blue) or warmer (up, ochre) the week of 6–12 January was than the week before. Single years are noisy: in 1953–1987 between 62% and 76% of winters had the second week colder; in 1988–2025, between 34% and 45% (a coin toss would give 50%). The thick line is the 15-winter mean.

The decline is gradual. Mean of the 19 stations over three periods: 1953–1974, −2.4 °C; 1975–1999, −0.5 °C; 2000–2025, +0.2 °C. Splitting at 1953–1987 and 1988–2025 was chosen after the data had been seen, so it is a way to describe the decline, not a date when it “stopped”. Winter 2026 is drawn but lies outside the periods counted.

#What 6 January has that other days do not

The same difference (the week starting on a day minus the week before) was computed for all 365 days of the year, 1953–1987. At 17 of 19 stations 6 January ranks 1st or 2nd, at the other two 4th. But the day in first place, at 17 stations, is 5 January, not 6; 3–9 January fill almost all the top places. It is one cold week counted several times, not a fixed date.

The step for each of the 365 days (week starting that day minus the week before), 1953–1987. 6 January in ochre.

Three caveats. First: 18 dates and calendars were tried in the search (nine feast days, each on the new and the old calendar), and over the whole 1950–2025 span in the ERA5 reanalysis 6 January ranks 3–4 of 365 at three of five Romanian points, but 29 at Cluj-Napoca and 112 at Timișoara; with 18 tests that does not survive correction on its own. The confirmation comes from independent stations, where the prediction was written before the count. Second: the other eight candidates (Dochia, St George, Armindeni, St Elijah, St Mary, Holy Cross, St Demetrius, St Nicholas) gave nothing beyond chance, but the direction of the effect was assigned here, not quoted from folklore collections, so they are not tests of their real sayings. Third: on the old calendar (19 January) there is no cold step: the difference is between +0.3 and +1.9 °C at every station.

#The 19 stations

Circles are the 19 stations used; the filled circle is the chosen one. High-mountain stations are left out (short or patchy records).

All 19 stations

Station	Winters	1953–1987	Rank of 6 Jan	1988–2025
Sibiu	27	−3.6 °C	2	+1.3 °C
Miercurea Ciuc	26	−3.3 °C	2	+1.5 °C
Ocna Șugatag	26	−3.2 °C	2	+1.1 °C
Bacău	26	−3.2 °C	2	+0.5 °C
Botoșani	26	−3 °C	2	+0.3 °C
Deva	26	−2.9 °C	2	+1.2 °C
Iași	29	−2.8 °C	2	+0.4 °C
Caransebeș	26	−2.8 °C	2	+1.1 °C
Cluj-Napoca	35	−2.7 °C	2	+1 °C
Călărași	35	−2.3 °C	2	+0.5 °C
Roșiori de Vede	26	−2.2 °C	4	−0.1 °C
Galați	29	−2.2 °C	2	+0.4 °C
Constanța	28	−2.1 °C	2	+0.6 °C
Craiova	26	−2.1 °C	1	+0.3 °C
Râmnicu Vâlcea	26	−2 °C	2	+0.9 °C
Buzău	35	−1.9 °C	2	+0.6 °C
București-Băneasa	35	−1.9 °C	2	+0.4 °C
Arad	35	−1.8 °C	4	+0.9 °C
Drobeta-Turnu Severin	35	−1.7 °C	2	+0.5 °C

#Not only a Romanian thing

To see whether the pattern is local, the same measurement was made on the ERA5 reanalysis (a grid, not stations) for five Romanian towns and four abroad. The 1953–1987 step appears at Vienna, Sofia and Warsaw as well; at Berlin it is weaker (−0.8 °C, rank 23 of 365). ERA5 smooths the weather, so its figures compare with each other, not with the stations’.

ERA5 point	1953–1987	rank of 365	1988–2025	rank of 365
Bucharest	−2.2 °C	1	+0.6 °C	326
Cluj-Napoca	−2.2 °C	3	+1.1 °C	362
Iași	−2.3 °C	2	+0.2 °C	243
Timișoara	−1.6 °C	3	+1.1 °C	361
Constanța	−1.9 °C	2	+0.6 °C	336
Vienna	−1.7 °C	3	+0.7 °C	334
Sofia	−1.7 °C	3	+0.7 °C	347
Warsaw	−1.3 °C	4	+0.7 °C	331
Berlin	−0.8 °C	23	+0.5 °C	313

What the table does not say: why. No meteorological explanation was tested here.

#The saying, and who looked before

The explanatory dictionary (DEX) gives “Gerul Bobotezei = ger mare (ca la început de ianuarie)”, a hard frost, as in early January. It does not say the frost comes after 6 January or is harder than on other days; the saying is about frost, not about a step between two weeks. This page measures only that step.

Churches that keep feasts on the old calendar celebrate Epiphany on 19 January. On 19 January there is no cold step (below).

In meteorology, typical weather on a date is called a “singularity”: a calendar-bound departure from the usual run of the year. Their existence is disputed; a 2002 paper asks whether North America’s “January thaw” is “a statistical phantom” (Godfrey, Wilks and Schultz). Alexander Buchan proposed in 1869 cold and warm spells each year from Scottish observations. For Central Europe there are singularity calendars (Flohn and Hess, 1949), and German sources list among them the “Christmas thaw” and the “sheep’s cold”. No dedicated climatological study of the Epiphany frost was found; the press treats it as an expression.

#How it was checked

The data are daily maximum and minimum temperatures at NOAA GHCN-Daily stations in Romania, without days flagged by NOAA’s quality control; the day’s mean is (max + min) / 2. The “normal” is the station’s mean for that calendar day over all years in its file, smoothed over 21 days. A winter’s step = mean departure over 6–12 January minus mean over 30 December–5 January, requiring at least four valid days in each week. 19 stations are used: the seven where the pattern was first seen and 12 chosen afterwards.

Before counting the 12 new stations, the date (6 January), the windows (seven days), the definition of normal, the periods (1953–1987 and 1988–2025) and the thresholds were written in a dated file: first-period step ≤ −1.0 °C, 6 January ranked ≤ 10 of 365, and second-period step > −0.5 °C, each at three quarters of the stations or more. Result: 12 of 12 for all three. This is the check that counts; ERA5 served only for the initial search and the comparison with other countries.

A sensitivity test: the normal might retain part of the cold week and inflate the positive post-1988 step. Redone with a normal that leaves out 27 December–15 January, and with one from three annual harmonics, the 1988–2025 step stays positive (+0.7 and +0.5 °C on average) and the 1953–1987 step (−2.6 and −2.7) stays as large.

Limits. NOAA states that GHCN-Daily series are not homogenised; a station move or urbanisation could shift absolute temperatures, but the step is a difference within one winter and is little affected. Some stations have only 26–29 winters in the first period (those starting in 1961 or later in the archive), and the count is shown at each. The mechanism was not tested. This is a description, not proof that a calendar date steers the weather.

Cite the NOAA data as: Menne et al. (2012), Global Historical Climatology Network - Daily (GHCN-Daily), Version 3, NOAA National Centers for Environmental Information, doi:10.7289/V5D21VHZ, version 3.35 (update of 6 November 2026), accessed 9 November 2026; and Menne et al. (2012), J. Atmos. Oceanic Technol. 29, 897–910. NOAA gives no warranty on the data. ERA5: Copernicus Climate Change Service, via Open-Meteo (CC BY 4.0).

#Questions

What do the measurements say about the Epiphany frost?

At weather stations in 1953–1987 the week of 6–12 January averaged, relative to normal, 2.5 °C colder than the week before, at all 19. In 1988–2025 no cold step shows. That is what the measurements say; they do not say why.

Does the weather follow the feast?

The measurements do not show that. It is a pattern described in temperature records, with 18 dates tried and a split into periods chosen after the data had been seen; the cause is not established.

Why is it gone after 1987?

The cause was not tested. The decline is gradual, not a collapse in 1988; that is all these data show.