The European model’s forecast, checked against 873 days

How often the weather forecast is wrong, from tomorrow to a week ahead

The day after tomorrow isn’t a coin flip: a rain forecast that far ahead comes true two times in three. Day five is the coin flip (on any given day it rains one day in five), and a week out 6 in 10 of the rains on the forecast never fall. The daily high misses by 1.5 °C on average for tomorrow and 2.6 °C a week ahead, measured here across Europe and, town by town, in Romania.

Pick a Romanian town How far the daily high misses there, from tomorrow to a week ahead: tap the map or search for the place.

How far the forecast high misses

tomorrow, gridtomorrow, station7 days, grid

How far the forecast high for tomorrow misses on average (dot: the E-OBS grid; ring: the weather station, where one is close enough) and a week ahead (square, grid).

#How accurate the forecast is in each Romanian town

The map shows 48 measured places: every county town (Buftea for Ilfov, the county around Bucharest), four mountain towns and two on the Black Sea coast. The colour shows how far, on average, the forecast high for tomorrow misses. Tap a county or a dot, or search for any place: one without a dot gets the nearest measured place, with the distance.

Pick a town on the map or search for a place to see how far the forecast high misses there.

Rain, the median of 28 county towns (the same on every card): forecast for tomorrow, it falls 68% of the time; five days ahead, 52%; a week ahead, 41%.

How far the high misses, by how many days ahead the forecast is made. Band: the middle half of the 42 county towns; line: their median.

All 48 places, as a table

Average miss of the high, in °C, against the E-OBS observation grid, most accurate first. Last column: the signed June–August miss at the weather station (minus = forecast too cool).
Place	Tomorrow	3 days	7 days	Days off by ≥3 °C, tomorrow	Summer, at the station
Mangalia	1.1	1.2	2.1	5%	+2.3 °C
Zalău	1.1	1.3	2.4	3%	−1.2 °C
Slatina	1.2	1.4	2.4	4%	−1.1 °C
Tulcea	1.2	1.4	2.4	6%	−1.6 °C
Constanța	1.2	1.4	2.3	6%	+1.4 °C
Craiova	1.3	1.6	2.6	6%	+0.1 °C
Timișoara	1.3	1.6	2.7	5%	−1.5 °C
Focșani	1.3	1.5	2.4	6%	−1.2 °C
Pitești	1.3	1.5	2.3	6%	−1.1 °C
Cluj-Napoca	1.4	1.5	2.5	6%	−2.2 °C
Târgu Mureș	1.4	1.5	2.4	9%	−1.9 °C
Râmnicu Vâlcea	1.4	1.6	2.4	6%	−2.2 °C
Brăila	1.4	1.6	2.5	5%	−1.7 °C
Iași	1.4	1.7	2.8	6%	−1.5 °C
Bistrița	1.4	1.5	2.4	7%	−2.5 °C
Slobozia	1.4	1.6	2.5	5%	−1.1 °C
Alexandria	1.4	1.7	2.5	7%	−1.5 °C
Vaslui	1.4	1.6	2.6	6%	−1.9 °C
Arad	1.4	1.6	2.7	6%	−2.0 °C
Galați	1.5	1.7	2.6	7%	−1.1 °C
Satu Mare	1.5	1.6	2.5	7%	−2.0 °C
Alba Iulia	1.5	1.6	2.6	9%	−2.1 °C
Bacău	1.5	1.7	2.9	9%	−1.2 °C
Ploiești	1.5	1.6	2.4	10%	−1.6 °C
Oradea	1.5	1.7	2.7	8%	−2.1 °C
Suceava	1.6	1.7	2.8	9%	−1.3 °C
Sibiu	1.6	1.7	2.6	9%	−1.2 °C
Botoșani	1.6	1.8	2.9	9%	−2.0 °C
Vatra Dornei	1.6	1.8	2.8	14%	—
Călărași	1.6	1.9	2.8	9%	−1.7 °C
Petroșani	1.6	1.8	2.6	14%	−2.7 °C
Buftea	1.6	1.8	2.6	8%	−1.2 °C
Buzău	1.7	1.8	2.6	12%	−2.2 °C
Giurgiu	1.7	1.9	2.6	11%	−1.7 °C
Sfântu Gheorghe	1.7	1.7	2.5	14%	−2.3 °C
Târgoviște	1.7	1.8	2.4	15%	−2.0 °C
Miercurea Ciuc	1.7	1.9	2.7	10%	−2.3 °C
Piatra Neamț	1.8	1.8	2.7	16%	−1.3 °C
Sulina	1.9	2.1	2.7	21%	+2.5 °C
Deva	1.9	2.0	2.9	16%	−2.1 °C
Baia Mare	1.9	1.9	2.7	16%	−3.2 °C
Bucharest	1.9	2.0	2.8	13%	−2.1 °C
Drobeta-Turnu Severin	2.0	2.0	2.8	18%	−2.6 °C
Brașov	2.0	2.0	2.5	21%	−1.9 °C
Reșița	2.0	2.1	3.0	19%	−2.6 °C
Sinaia	2.0	2.1	2.7	24%	—
Târgu Jiu	2.1	2.2	2.9	24%	−3.1 °C
Predeal	2.2	2.2	2.7	28%	−1.7 °C

#Day by day: the forecast against the station thermometer

Each bar is one day between 8 February 2024 and 24 August 2025: how far the forecast high was from the high measured at the town’s weather station. Below the line the forecast was too cool; above it, too warm. Choose how many days ahead the forecast was made, and move over the bars to see the day.

Forecast made:

Pick a town on the map to see its days.

#The day after tomorrow isn’t a coin flip. Day five is

The day after tomorrow isn’t a coin flip. Day five is.

Of every 10 days the European model forecast rain (1 mm or more), how many it actually rained. Median of 28 Romanian county towns, February 2024 – June 2026.

In the 28 Romanian county towns counted here it rains (1 mm or more) one day in five: 21% of days. When the European model shows rain for tomorrow, it actually rains 68% of the time. For the day after tomorrow, 65%; three days ahead, 60%. Five days ahead, 52%: a coin flip, but still 2.5 times as often as on any given day. A week ahead, 41%.

Forecast made	Forecast rain actually falls	Of the rain that fell, share forecast	Days right (rain or not)	Skill over chance (0 = guessing)
for tomorrow	68%	83%	89%	0.67
for the day after tomorrow	65%	79%	87%	0.62
3 days ahead	60%	74%	85%	0.56
5 days ahead	52%	67%	81%	0.46
a week ahead	41%	57%	74%	0.30

Median of 28 Romanian county towns, February 2024 – June 2026. A rainy day = 1 mm or more in 24 hours. Saying “no rain” every day would be right on 79% of days.

The 14 county towns left out (Arad, Bacău, Baia Mare, Bistrița, Miercurea Ciuc, Oradea, Piatra Neamț, Pitești, Reșița, Satu Mare, Suceava, Timișoara, Târgu Mureș, Zalău) could not be aligned with confidence: in the observation grid their rain best matches a different 24-hour window from the rest of the country, and on the wrong window the forecast would look worse than it is.

A week out there is a paradox. Saying “no rain” every single day would be right on 79% of days. The seven-day forecast is right on only 74%, worse than “no rain” in all 28 towns (at five days, in 8 of them). That doesn’t make it useless: rain on a seven-day forecast still falls 2.0 times as often as rain picked at random. It is just that, from that far away, the forecast shows rain on more days than it rains: 27% of days, against 21%.

Summer is harder. Of the rains forecast for tomorrow in June–August, 46% never fall; in winter, 26%. The score is lower in summer in 28 of 28 towns. A possible reason: summer showers are smaller than the model grid, and sometimes smaller than the station network.

There is no town ranking for rain. Rain forecast for tomorrow fell between 60% and 76% of the time depending on the town, but the order of towns in 2024 does not repeat in 2025–2026 (the correlation between the two periods is 0.08, practically zero). What repeats is the fall with each day ahead, so every town card shows the median of the 28 county towns for rain.

#In summer the forecast high runs too cool almost everywhere inland

In June, July and August the forecast high for tomorrow was lower than the measured one at 42 of 43 inland Romanian stations, by 1.9 °C on average (the stations’ median). Most at Baia Mare (3.2 °C) and Târgu Jiu (3.1 °C). The one exception is Craiova.

forecast too coolforecast too warmcircle size: how many degrees

How far the forecast high for tomorrow was below (blue) or above (red) the station’s measured high, on average: June–August 2024 and June to 24 August 2025.

On the Black Sea coast it is the other way round: the forecast ran 1.4 °C too warm at Constanța, 2.3 °C at Mangalia and 2.5 °C at Sulina. A possible reason, not tested here: a model grid cell, about 25 km across, mixes land with sea and misses the sea breeze.

The E-OBS observation grid, built from other measurements, shows the same sign in all 43 inland towns, in summer 2024 and after. For inland towns this data does not say why. Part of it may come from how the forecast high is read here, as the warmest forecast hour, while a thermometer also catches the peak between hours; this data cannot say how much.

#The forecast high in 16 European cities and Bucharest

The same forecast, for the same days, in 16 large European cities (daily high only: rain is recorded differently from one country to the next). Bucharest is among the three least accurate both on the observation grid, of the 17 cities (1.9 °C for tomorrow), and at its station, of the 14 with a station close enough (2.1 °C).

The rest of the order depends on the source. Athens looks the most accurate on the grid (0.9 °C), but at its station, 11 km away on the coast, the miss is 1.8 °C. London is the reverse: 2.1 °C on the grid, 1.6 °C in St James’s Park. So the chart shows both sources and names no winner.

#Questions

How reliable is a 7-day weather forecast?

A week out, the forecast high misses by 2.6 °C on average, and one day in three is off by 3 °C or more. Rain forecast seven days ahead falls 41% of the time, against 21% on any given day.

Is the forecast for the day after tomorrow a coin flip?

No. Rain forecast for the day after tomorrow falls 65% of the time, two times in three. Day five is the coin flip (52%), while on any given day it rains one day in five (21%).

Why does the forecast high run too cool in Romanian summers?

It did at 42 of 43 inland stations, by 1.9 °C on average; on the coast it runs warm. On the coast a possible reason is the model cell mixing land and sea. Inland, this data shows the effect, not the cause.

Is this about the weather app on my phone?

Not directly. These are scores for the European ECMWF IFS model’s raw forecast, as kept in the Open-Meteo archive. Apps often start from the same models, then correct or blend them in their own ways.

Which Romanian towns get the most accurate forecast high?

Slatina and Craiova are among the six most accurate county towns on both sources; Târgu Jiu and Reșița among the three least accurate. The order of towns repeats from one year to the next (correlation 0.72 between 2024 and 2025–2026).

#How it was measured

The forecast: the European ECMWF IFS model at 0.25°, from Open-Meteo’s archive of earlier forecasts, at each town’s point. For every hour the archive keeps the value forecast 24, 48, 72, 120 and 168 hours earlier; here that is the forecast for tomorrow, the day after, and 3, 5 and 7 days ahead. The archive for this model starts in February 2024. One fixed model: the archive’s “best model” option silently switches models from one year to the next.

For every hour of the day (00–24 UTC) the value forecast exactly 24 (or 48, 72...) hours earlier is taken; the forecast high is the warmest of them. Forecast rain is the total over the 24 hours ending at 06 UTC, the window that best matches the observation grid in most Romanian cells.

What actually happened comes from two independent sources. The E-OBS v34.0e observation grid (0.1°, nearest cell), 873 days, 8 February 2024 – 29 June 2026, for highs and rain. And the nearest weather station in NOAA’s GSOD archive, for highs: 553 days, 8 February 2024 – 24 August 2025 (the archive has no later Romanian data yet). A station is used only within 15 km and 150 m of height of the forecast point, which leaves out Sinaia (its station is at 1,510 m) and Vatra Dornei (Poiana Stampei, 17 km). The same rule drops Berlin, Budapest and Munich’s stations.

Scores: the average miss of the high (degrees, either way), the share of days off by 3 °C or more, and the signed miss in summer; for rain, yes/no tables and the Heidke skill score national weather services use on their own forecasts (0 means no better than guessing with the same rain frequency). The tests were written before the counting. The forecast beats both naive forecasts: the 2014–2023 average for the date misses by 3.6 °C, “same as today” by 2.4 °C for tomorrow and 4.7 °C a week out.

What held up: the order of towns for the daily high repeats between 2024 and 2025–2026 (correlation 0.72) and agrees between grid and stations (0.56). What did not: a town ranking for rain (correlation 0.08), and Brașov as the least reliable town, which comes out near the bottom on the grid but mid-table at its Ghimbav station.

These are scores for the European model’s raw forecast at the town’s point, not for any particular app. The page does not publish E-OBS daily values, only scores; the days shown one by one come from stations (public domain) and the archived forecast (CC BY 4.0).

#Sources

Acknowledgement required by ECA&D: “We acknowledge the E-OBS dataset and the data providers in the ECA&D project (https://www.ecad.eu).” Cornes, R., G. van der Schrier, E.J.M. van den Besselaar and P.D. Jones, 2018, J. Geophys. Res. Atmos., 123. Forecasts: Open-Meteo.com, CC BY 4.0.