Data sources and licences

Wind gusts at 10-minute, hourly and daily cadence; precipitation at 10-minute, hourly and daily cadence; hourly present-weather observations and daily observed phenomena including hail. Published at opendata.dwd.de and reused here under CC BY 4.0.

The archive is published in two tiers, and the difference matters for a document: the historical archive is frozen and its quality control is complete, while the rolling recent archive covers roughly the last 500 days and is published before quality control finishes, so those values can still change. Every record states which tier each figure came from.

Attribution as required: Quelle: Deutscher Wetterdienst. Datenbasis: Deutscher Wetterdienst, Werte nach Ort und Zeitraum ausgewählt und in km/h umgerechnet.

For addresses in Austria: wind gusts and precipitation at 10-minute cadence from the national station archive (klima-v2), 1992 to yesterday, published at data.hub.geosphere.at and reused here under CC BY 4.0. Hourly values supply the ten-year comparison. The archive holds no present-weather observations, so a record for an Austrian address states that no instrument observed hail — not that there was none.

Every value carries the service's own quality indicator. A value the service has checked, automatically or by hand, may decide the verdict; one it has not checked is printed and marked but decides nothing. Where the service corrected a value, the record says so.

Attribution as required: Datenquelle: GeoSphere Austria — https://data.hub.geosphere.at

For addresses in Poland: hourly synoptic station observations from the national archive, 2001 to the last closed month, published at danepubliczne.imgw.pl. Wind gusts come from the archive's period-maximum field, which carries the minute the gust actually blew; precipitation from the six-hour totals reported at the main synoptic terms; hail from the station's own present-weather observation. The ten-year comparison is built from the same archive, one file per station per year.

The period maximum does not exist before 2011: in those years the archive publishes only the gust measured at the observation itself, so a record for such a date says which of the two columns it read, and the ten-year comparison is measured with that same column. A year the column is not published in is left out of the window entirely rather than counted as calm, and where what remains is under a year of days no rank is printed at all. A published value is not automatically a sound one either: where one row of the archive reports a gust that another row of the same archive says did not occur, neither is printed as a fact and the contradicted value is dropped.

The archive is published about five working days after a month closes, and only after the institute's own verification — so a value in it has been checked, and there is no separate provisional tier to warn about. The institute does state that corrections can still be issued, mostly for the current year, and it lists every one of them in a public change log, which each record cites alongside the data files.

The archive does not publish station coordinates with the measurements, so the distance from an address to a station is computed from the NOAA station history, joined on the station's WMO index. Where a station cannot be placed that way it is excluded from the record rather than positioned by guesswork. A quarter of the network is placed from a coordinate that history publishes to a tenth of a degree — a few kilometres of slack at Polish latitudes — and a record quoting such a station says so, because the distance printed beside it is finer than the placement behind it.

The institute's terms are published at danepubliczne.imgw.pl/pl/regulations, and every record links to them. They reserve at § 6 ust. 4 that published data may be unverified; the description accompanying this particular catalogue states that its data are verified, which is why a record from it says so — the general reservation is why each record also cites the change log and can be re-checked against the archive itself, in public, at any time.

Attribution as required by the institute's terms (§ 5): Źródłem pochodzenia danych jest Instytut Meteorologii i Gospodarki Wodnej – Państwowy Instytut Badawczy. Dane Instytutu Meteorologii i Gospodarki Wodnej – Państwowego Instytutu Badawczego zostały przetworzone.

For addresses in the Netherlands: hourly station observations from the national archive, published at daggegevens.knmi.nl and reusable under CC0. Wind gusts come from the archive's FX field — the strongest gust of each hourly division — precipitation from the hourly sum RH, and present weather, where the archive carries it, from WW. The ten-year comparison is built from the same hourly endpoint, so both ends of it are measured the same way.

The archive files every value under a UTC hour, and a Dutch calendar day runs one or two hours ahead of that, so a day here is assembled from the hourly divisions that begin inside it — including the last ones filed under the previous UTC date. At De Bilt over the ten calendar years 2016 to 2025 this changes the day's strongest gust on 196 of 3,653 days, and on 104 of them the peak belongs to an hour filed under the day before. The archive gives no minute, only the hour, so the time printed beside a gust is the whole hour it fell in, in local time, and never a minute the archive does not hold.

KNMI checks these values during office hours. The institute states that the most recent data may still be unvalidated, and that the validated series usually appear a day or a few days after the measurement, in some cases up to two weeks. There is no per-value quality flag to print, so a record for a date inside that window says the archive is still provisional, and one for a date outside it says the archive is closed and carries the service's own control.

Present-weather observations stop at the end of 2020: from 1 January 2021 the weather-code column is empty at every station. A record for a Dutch address on a later date therefore states that no instrument observed hail — which is not a statement that no hail fell. For dates up to the end of 2020, hail is reported where a station's own present-weather code says so.

Attribution as required: bron: KNMI. The KNMI is not involved in this derived work and does not necessarily endorse its contents.

For addresses in Czechia: ten-minute station observations from the national archive at opendata.chmi.cz, free to use under Creative Commons Attribution 4.0, as the institute's own FAQ states. Wind gusts come from the Fmax field — the strongest gust of each ten-minute interval — precipitation from SRA10M, and hail from the station's own record of observed phenomena, where the code KR means hail. The comparison with earlier years is built from the same ten-minute files, reduced to a daily maximum on the same local-day rule the day itself is read with.

The archive stamps each value with the moment its ten-minute interval begins, in UTC, and a Czech calendar day runs one or two hours ahead of that. A day here is therefore assembled from the intervals that begin inside it, including the last ones filed under the previous UTC date. The archive publishes no minute inside the interval — it has a field for the time of the maximum, and the institute nowhere states what its numbers mean — so the time printed beside a gust is the ten minutes it fell in, in local time, and never a minute the archive does not hold.

Every value carries the institute's own quality code and the record prints its meaning. Only a value the institute marks good decides anything: the strongest-gust figure, the statement about the threshold and the one derived number come from those alone. A value marked estimated, suspect, of low quality or of unknown quality is printed because the archive holds it, and labelled; a value marked missing is not a reading at all. The checking runs behind the measurement — in the file for 1 September 2026, rebuilt at 12:11 the next day, every value up to 12:10 carried the good code and every value after it carried unknown — so a record for a very recent date often states its figures and withholds the verdict on the threshold, and the same date read again after the month closes carries one.

The archive is published in tiers, and the record says which one each figure came from. The current day is rewritten every hour and a day's file is finished shortly after midnight UTC the next morning; the rolling files are rebuilt twice a day after checks and corrections; the monthly files appear on the third day of the following month; and the frozen yearly files — ten-minute data for every automated station since 2018 — are updated once a year, by the end of February. That comparison is taken from the frozen tier only, so that the figure being ranked and the series it is ranked against are measured to the same standard; and because the ten-minute archive itself begins in 2018, the window is shorter than ten years, so the record names the years it actually read rather than promising a decade. On the first two days of a month, before the previous month's own file is uploaded on the third, only its last dozen days are still in the per-day files — a date earlier in that month is answered from the international hourly network instead, and the record says so.

Attribution as required: Zdroj dat: Český hydrometeorologický ústav (ČHMÚ), opendata.chmi.cz, CC BY 4.0.

For addresses in Switzerland: measurements from the SwissMetNet automatic network, published station by station at data.geo.admin.ch and reusable under CC BY 4.0. Wind gusts come from the archive's own gust-peak field — fkl010z1 for a ten-minute interval, fkl010h1 for an hour — precipitation from rre150z0 or rre150h0. The ten-minute series begins on 1 February 2004 at most stations and the hourly one in 1981, so a recent date is answered to the ten minutes and an older one to the hour, and the record says which, because the interval it prints differs.

Every reference time stamp in this archive is in UTC, and for the ten-minute and hourly values the stamp is the end of the measurement interval — the weather service states both, and states one exception to the second: hourly values before 2018 were built on the synoptic schedule. Such an hour is the strongest of the six ten-minute values stamped from ten minutes before it to forty minutes after it, so it covers the sixty minutes from twenty minutes before the hour to forty minutes after it. It is a full hour like any other and consecutive ones follow each other exactly; what they do not do is break at the hour, so they never line up with a local day. A Swiss calendar day runs one or two hours ahead of UTC, and a day here is assembled from every interval that overlaps it. In that older era the two intervals at the ends of a day straddle midnight and belong to both dates they touch, so where a printed interval begins on the previous evening the record prints its date beside it. There is no minute inside an interval to print, so the time beside a gust is the interval itself, written in local time with both ends — and because those older hours cannot be laid on a local day, a record that reads them states no day rainfall total.

The weather service publishes no quality flag beside an individual value. What it does say is that ground-based measurements are initially raw, that it checks their plausibility over a rolling period of five days from the time of measurement, and that the manually checked data are normally published five days after the measurement. So a record for a date inside those five days says the archive is still provisional; one for an older date in the current year says the archive carries the service's own control; and one for a closed year says the archive is frozen. No figure is ever labelled unchecked, because the service has not said that about any individual reading.

Hail comes from a different collection of the same service: the meteorological visual observations, made by a person between two and eight times a day, every day of the year, at about twenty places in the country. The daily parameter is w3p002d0, 'day with hail'. A record therefore cites the nearest observing station's yes or no for the whole date, with the distance to it, and says plainly that this is a human observer's daily observation and not an instrument reading at the address. A blank in that column is not a no — it means no observation is recorded — and such a station is not counted among those that reported nothing. The service's radar hail products are kept for fourteen days and only in the summer half-year, so they cannot evidence a past date and are not read.

One gap, said plainly. The frozen decade files are rebuilt once a year, and the year that has just closed does not appear in them until February: the 2020–2029 hourly file for Zürich/Kloten ended on 31 December 2025 and was rebuilt on 18 February 2026, while the rolling file begins on 1 January of the current year. So for about seven weeks each New Year, a Swiss date in the previous year is in no Swiss file at all. A check for such a date is answered from the international hourly network instead — the same archive that answers everywhere no national one reaches — and the record names that network's station and files in the usual way. Once the annual rebuild lands, the Swiss archive answers for that date again.

Attribution as required by the terms of use: Quelle: MeteoSchweiz · Source: MétéoSuisse · Fonte: MeteoSvizzera · Source: MeteoSwiss. Acknowledging the source does not make it appear that MeteoSwiss supports this document or its use.

For addresses in Spain, including the Balearics, the Canaries, Ceuta and Melilla: the daily climatological values of the national station network, published through AEMET's open-data API at opendata.aemet.es. Wind gusts come from the archive's racha field — the strongest gust of the day — with horaracha, the hour and minute it blew; precipitation from prec and mean wind from velmedia. The archive holds no present-weather field, so a record for a Spanish address states that no instrument observed hail — not that there was none. The comparison with earlier years is built from the same daily values, reduced on the same local-day rule the day itself is read with. This archive is read for Spanish addresses only: its day is the UTC day and its clock is Spain's, so a house in Portugal or Morocco — which a Spanish instrument may well stand nearest to — is answered by the international hourly network instead, which carries its own instant with every observation.

The archive publishes one gust for each station and each day, and its day is the UTC calendar day, while Spain's clock runs one or two hours ahead of it — in the Canaries one hour, and only in summer. A local day is therefore assembled from the UTC days whose stamped minute falls inside it, which means reading the day before as well: a gust at 23:30 UTC belongs to the following local day and is printed under it. Where the archive gives no usable minute — it marks the time «Varias», meaning the maximum was reached several times, or the field is empty, or it holds something that is not a time on the clock, such as the «65:50» it publishes at one Canary station — there is no minute to print, so the figure is stated as the whole UTC day it belongs to, written in local time. The record says which of those it was, and never puts the word «Varias» on a value the archive did not mark that way. One thing this archive cannot show, and the record says so: a gust that was not the strongest of its own UTC day is not published at all, at any hour.

Precipitation in this archive is measured from 07 to 07 UTC and not over the local calendar day, so the figure printed is the archive's own daily total filed under the date, and no twenty-four-hour local total is stated. The archive's own marks are printed as marks: «Ip» is precipitation the gauge measured below its 0.1 mm resolution, and «Acum» is an accumulation over several days, which is not this day's rainfall and is not given as one.

AEMET publishes these values once a day with a stated delay of four days, and publishes no quality indicator beside an individual value. Its own metadata for the product names the generating unit, the cadence, a description and the licence, and says nothing about checking, validating or revising a value — so a record states the publication schedule and stops there, rather than crediting the service with a control it has never claimed. Reaching the archive needs a free API key, which AEMET issues at opendata.aemet.es and which is valid for three months; every record cites the exact request that returns the row behind each figure, along with the row's own date and station code, so anyone with a key of their own can fetch the same row. The key is sent as a request header and never appears in a citation. Without a key the cited request answers with an empty body rather than an error, which is the service's own behaviour; the document says so where it names the address, so that nobody mistakes it for a citation that was invented.

Attribution as required by the archive's own copyright statement: © AEMET. Autorizado el uso de la información y su reproducción citando a AEMET como autora de la misma. Commercial and non-commercial reuse is permitted under Ley 18/2015; see aemet.es/es/nota_legal.

German postal places and the districts that own them, used to resolve a postal code when an address cannot be matched, and to name the district a record is about. Licensed CC BY 4.0.

Address geocoding, via the Nominatim service. © OpenStreetMap contributors, ODbL. The precision the geocoder achieved is printed in every record, because a house number and a postal-code centre are not the same claim.

Lightning. There is no lightning network in Europe whose data can be licensed for this at a price that fits a €19 record, and the community networks that publish freely forbid exactly this use in their own terms. So lightning is not reported at all.

Radar-derived hail for past dates. The open radar archive of reflectivity and hail classification is kept for 48 hours. It cannot be reconstructed afterwards, and the deep gridded archives hold rainfall in millimetres, which says nothing about hail. Records therefore state hail only where a station observed it.