Time zones and reference times#

wetterdienst returns every timestamp as datetime[us, UTC], the same type for every provider and for values, interpolated values and summaries; the provider tests hold the frame of every result they build to it. Providers do not publish that way. They stamp in UTC, in local civil time, in local standard time or in a fixed offset, and they stamp the start, the end or the middle of the period an aggregate covers. A mistake is an hour off, or a day at a daily boundary, and never an error, so this page records what each provider’s own documentation says and what the library does with it.

How to read the tables:

  • source time zone: the zone in which the provider states its stamps.

  • reference time: where the stamp sits in the period a value covers. instant is a reading at the stamp, start and end an aggregate over the period that begins or ends at it, middle one centred on it, and mixed that the datasets, parameters or periods of the row differ (the qualifier says how). A daily stamp is mostly a date at 00:00 UTC, and for it start and end say that the day’s window begins or ends on that date, and the qualifier gives the window.

  • unverified means the provider’s documentation does not state it. The cell never holds a guess; where a comparison with a second source suggests an answer, the cell says so in brackets as an observation.

  • verified: yes if the quoted text defines both zone and reference time as shown, partly if it defines only one of them or one rests on the stamp’s format or the field names alone, no if the documentation defines neither (both cells are then unverified).

  • The notes list, per row, what the library does with the stamp and the sentences of the provider the row rests on. German, French, Polish and Spanish quotations are given in the original.

A daily stamp is the date of the day it belongs to, at 00:00 UTC, for most providers, whatever window the provider defines for that day (06:00 to 06:00 UTC for precipitation is common) and whatever zone the provider keeps its days in, so that a request for one day (start and end the same date, which are 00:00 UTC) returns that day’s values where it is so (GH-2683). A monthly stamp is the first of the month at 00:00 UTC. The notes say where it is not.

Rows with a confirmed deviation from their source link the issue that tracks it, and the issue is:

  • #3029: hourly values stamped in MEZ that no Zusatz-Info marks: the soil temperature of the eMD periods and the air temperature of Diepholz from 1995-02-01 to 1996-05-31 were returned one hour late and are shifted to UTC with zone:mez; the plain MEZ periods of Regensburg-Oberhub and Stötten change their clock inside the label and are not settled, and the hourly wind at Bückeburg and Celle is right as written (NOAA ISD is the one an hour off there)

  • #2711: the MEZ and MOZ periods of the subdaily and daily observations were compared with NOAA ISD and GHCN-daily on samples (the notes of dwd give the results): the MEZ label of 1973 to 2000 is no whole-hour shift, the MEZ label up to 1946 and the MOZ label up to 1972 look real but fit no whole hour reliably, and no daily stamp is a day off

  • #2816: the hourly sun stamps of the periods DWD calls Stundenwerte in WOZ (local true solar time, not UTC) are returned as written, unshifted, and the rows are flagged zone:woz in quality_flags

A deviation that is suspected but was not reproduced, for want of credentials or of a second source, says so in the note of its row.

The type check on the timestamp column of every values result runs in each provider test (tests/provider/conftest.py).

aemet#

network

resolution

datasets

source time zone

reference time

verified

observation

hourly

all

UTC

mixed (end of the stamped hour: instant for ta/tpr/hr/pres/pres_nmar; end of the preceding 60 min for tamax/tamin/prec/vmax/dmax; end for wind: the mean of the last 10 minutes before the stamp (vv/dv))

yes

observation

daily

data

unverified (prec day ‘07 a 07’ with no zone named)

unverified (prec: day 07-07; which date ‘fecha’ carries not stated)

no

observation

daily

data

unverified (only the hora* times of day are stated as UTC)

unverified (stamp is a calendar date; day boundaries not stated)

no

observation

monthly

all

unverified

unverified (stamp names a calendar month ‘AAAA-X’)

no

observation

annual

all

unverified

unverified (stamp names a calendar year, ‘AAAA-13’)

no

Notes:

  • observation/hourly: Library: src/wetterdienst/provider/aemet/observation/api.py parses ‘fint’ with ‘%Y-%m-%dT%H:%M:%S%z’ and converts to UTC; no shift. The endpoint takes no date range (api.py, rolling window); afterwards src/wetterdienst/model/values.py keeps rows with start <= timestamp <= end (closed both), naive start/end tagged and aware ones converted to UTC at src/wetterdienst/model/request.py. Note: the metadata gives ‘fint’ as ‘AAAA-MM-DDTHH:MM:SS’ without an offset; a stamp without one fails this format (checked: InvalidOperationError) - live format not seen, no API key. Source: fint: “Fecha hora final del período de observación, se trata de datos del periodo de la hora anterior a la indicada por este campo (hora UTC)” [end date-time of the observation period; the data are of the period of the hour before the one this field gives (UTC hour)], tipo_datos “string (AAAA-MM-DDTHH:MM:SS)”. ta: “Temperatura instantánea del aire correspondiente a la fecha dada por ‘fint’”. prec: “Precipitación acumulada, medida por el pluviómetro, durante los 60 minutos anteriores a la hora indicada por el período de observación ‘fint’”. vv: “Velocidad media del viento, media escalar de las muestras adquiridas cada 0,25 ó 1 segundo en el período de 10 minutos anterior al indicado por ‘fint’”. tamax: “valor máximo de los 60 valores instantáneos de ‘ta’ medidos en el período de 60 minutos anteriores a la hora indicada por el período de observación ‘fint’”. https://opendata.aemet.es/opendata/sh/55c2971b (AEMET metadatos, observacion/convencional; Spanish)

  • observation/daily (data): Library: Applies to prec (precipitation_amount). src/wetterdienst/provider/aemet/observation/api.py parses ‘fecha’ (‘%Y-%m-%d’) to 00:00 and replace_time_zone(‘UTC’): the stamp is 00:00 UTC of ‘fecha’, 7 h before a 07-07 window that starts on ‘fecha’ (or 17 h after one that ends on it). Request bounds are converted to UTC (src/wetterdienst/model/request.py) and sent as ‘%Y-%m-%dT%H:%M:%SUTC’ in 179-day chunks (api.py); afterwards src/wetterdienst/model/values.py keeps rows with start <= timestamp <= end (closed both), naive start/end tagged and aware ones converted to UTC at src/wetterdienst/model/request.py. Source: prec: “Precipitación diaria de 07 a 07” [daily precipitation from 07 to 07]; fecha: “fecha del dia (AAAA-MM-DD)” [date of the day]. The zone of ‘07’ and whether ‘fecha’ is the day the window starts or ends on are not stated (the mensualesanuales metadata states its wind run as “Recorrido medio diario (de 07 a 07 UTC)”, which is a different field). https://opendata.aemet.es/opendata/sh/b3aa9d28 (AEMET metadatos for climatologicos/diarios; this link is the ‘metadatos’ of the diarios example call in https://opendata.aemet.es/centrodedescargas/docs/FAQs170621.pdf p.26; Spanish); also searched the OpenAPI spec https://opendata.aemet.es/AEMET_OpenData_specification.json and the FAQ PDF for UTC/07: nothing more.

  • observation/daily (data): Library: Applies to all parameters except prec (tmed, tmax, tmin, dir, velmedia, racha, presmax, presmin, hrmedia, hrmax, hrmin). src/wetterdienst/provider/aemet/observation/api.py turns ‘fecha’ into 00:00 UTC of that date (start of a UTC day); request handling as for prec (api.py); afterwards src/wetterdienst/model/values.py keeps rows with start <= timestamp <= end (closed both), naive start/end tagged and aware ones converted to UTC at src/wetterdienst/model/request.py. Source: fecha: “fecha del dia (AAAA-MM-DD)”; tmax: “Temperatura Máxima del día”; horatmax: “Hora y minuto de la temperatura máxima” with unidad “UTC” (same ‘UTC’ unidad for horatmin, horaracha, horapresmax, horapresmin, horahrmax, horahrmin, horapintmax). The day’s boundaries/zone themselves are not stated. https://opendata.aemet.es/opendata/sh/b3aa9d28 (AEMET metadatos for climatologicos/diarios; this link is the ‘metadatos’ of the diarios example call in https://opendata.aemet.es/centrodedescargas/docs/FAQs170621.pdf p.26; Spanish)

  • observation/monthly: Library: src/wetterdienst/provider/aemet/observation/api.py extracts year and X from ‘fecha’ (^YYYY-M{1,2}$), keeps X != 13 and builds the 1st of month X; api.py casts to Datetime UTC (00:00 UTC, start of the month). Years requested come from the UTC-converted bounds (request.py; api.py, 3-year chunks); afterwards src/wetterdienst/model/values.py keeps rows with start <= timestamp <= end (closed both), naive start/end tagged and aware ones converted to UTC at src/wetterdienst/model/request.py. Source: fecha: “año y mes (AAAA-X) donde X es un número del 1 aa 13, indicando el mes y el valor 13 indica valor anual” [year and month (AAAA-X) where X is a number from 1 to 13 giving the month, 13 meaning the annual value]; p_mes “Precipitación total mensual/anual”; w_rec “Recorrido medio diario (de 07 a 07 UTC) del mes/año”. No zone for the month boundaries or for the daily values summed into p_mes. https://opendata.aemet.es/opendata/sh/997c0034 (AEMET metadatos, climatologicos/mensualesanuales; Spanish)

  • observation/annual: Library: src/wetterdienst/provider/aemet/observation/api.py keeps fecha ‘YYYY-13’ and builds January 1 of YYYY; api.py casts to Datetime UTC (00:00 UTC Jan 1, start of the year). Years from the UTC-converted bounds (request.py; api.py); afterwards src/wetterdienst/model/values.py keeps rows with start <= timestamp <= end (closed both), naive start/end tagged and aware ones converted to UTC at src/wetterdienst/model/request.py (so start=2024-06-01 drops the 2024 annual row). Source: fecha: “año y mes (AAAA-X) donde X es un número del 1 aa 13, indicando el mes y el valor 13 indica valor anual”. No zone for the year boundaries. https://opendata.aemet.es/opendata/sh/997c0034 (AEMET metadatos, climatologicos/mensualesanuales; Spanish)

chmi#

network

resolution

datasets

source time zone

reference time

verified

observation

10_minutes

all

UTC

unverified (observed: end; T the value at the stamp)

partly

observation

hourly

all

UTC

unverified (observed: end, the hour before the stamp)

partly

observation

daily

all

UTC

unverified (observed: SRA the 24 h from 06 UTC of the stamp’s date, TMA/TMI the 24 h to 20 UTC of the stamp’s date)

partly

observation

monthly

all

UTC

unverified (observed: start (library-built: YEAR-MONTH-01))

partly

observation

annual

all

UTC

unverified (observed: start (library-built: YEAR-01-01))

partly

Notes:

  • observation/10_minutes: Library: DT parsed ‘%Y-%m-%dT%H:%MZ’ and labelled UTC unchanged (chmi/observation/parser.py); month files (<prefix>-<WSI>-<el>-YYYYMM.csv) chosen from start/end converted to UTC first (api.py). Request bounds: naive -> UTC (model/request.py); output cut inclusively to [start, end] (model/values.py). Source: Czech, section “Adresář now” (the historical_csv section has no time statement): “V čase 00:02 byl 18. 6. 2024 vytvořen poslední soubor za předchozí den (17. 6. 2024) a data pro 18. 6. 2024 jsou postupně každou hodinu přepisována, v příkladu s časem 05:02 obsahují data od 00:00 do 04:50. Časy jsou v UTC.” [At 00:02 on 18.6.2024 the last file for the previous day was created and the data for 18.6.2024 are overwritten hourly; in the example at 05:02 they hold data from 00:00 to 04:50. Times are in UTC.] https://opendata.chmi.cz/meteorology/climate/Klimatologicka_data_popis.pdf. The CSV stamps carry a ‘Z’ (e.g. 2024-10-01T00:00Z). Searched the PDF (all 8 pages), metadata/meta1-4.csv: nothing on whether a stamp is the start or end of its interval nor on the day definition. Data check (not documentation), Cheb 0-20000-0-11406, Oct 2024: a month file holds stamps 01T00:00Z..31T23:50Z (UTC month); hourly SRA1H at H equals the sum of SRA10M stamped H-50..H (87/88 hours), consistent with 10-min stamps marking the end of the interval. Air temperature of Praha-Libuš 11520 against NOAA GHCNh Praha-Libuš, 212 full hours of 2025-06-01 to 10: mean absolute difference 0.045 K with the library’s stamp, 0.22 K and 0.24 K with it moved by -10 and +10 minutes, so the value is the instant at the stamp.

  • observation/hourly: Library: As 10_minutes: DT labelled UTC unchanged (parser.py), UTC month files (api.py). Request bounds: naive -> UTC (model/request.py); output cut inclusively to [start, end] (model/values.py). Source: Czech, section “Adresář now” (the historical_csv section has no time statement): “V čase 00:02 byl 18. 6. 2024 vytvořen poslední soubor za předchozí den (17. 6. 2024) a data pro 18. 6. 2024 jsou postupně každou hodinu přepisována, v příkladu s časem 05:02 obsahují data od 00:00 do 04:50. Časy jsou v UTC.” [At 00:02 on 18.6.2024 the last file for the previous day was created and the data for 18.6.2024 are overwritten hourly; in the example at 05:02 they hold data from 00:00 to 04:50. Times are in UTC.] https://opendata.chmi.cz/meteorology/climate/Klimatologicka_data_popis.pdf. The CSV stamps carry a ‘Z’ (e.g. 2024-10-01T00:00Z). Searched the PDF (all 8 pages), metadata/meta1-4.csv: nothing on whether a stamp is the start or end of its interval nor on the day definition. Data check (not documentation): SRA1H stamped H = sum of the 10-min SRA10M stamped H-50..H in 87 of 88 wet hours (start-labelled reading: 11), i.e. the hour ending at H.

  • observation/daily: Library: Row picked by TIMEFUNC (AVG/06:00/20:00/00:00 per element, api.py); DT labelled UTC and truncated to 00:00 of its UTC date (parser.py), e.g. SRA ‘2024-10-02T06:00Z’ -> 2024-10-02 00:00 UTC. Request bounds: naive -> UTC (model/request.py); output cut inclusively to [start, end] (model/values.py). Source: Czech, section “Adresář now” (the historical_csv section has no time statement): “V čase 00:02 byl 18. 6. 2024 vytvořen poslední soubor za předchozí den (17. 6. 2024) a data pro 18. 6. 2024 jsou postupně každou hodinu přepisována, v příkladu s časem 05:02 obsahují data od 00:00 do 04:50. Časy jsou v UTC.” [At 00:02 on 18.6.2024 the last file for the previous day was created and the data for 18.6.2024 are overwritten hourly; in the example at 05:02 they hold data from 00:00 to 04:50. Times are in UTC.] https://opendata.chmi.cz/meteorology/climate/Klimatologicka_data_popis.pdf. The CSV stamps carry a ‘Z’ (e.g. 2024-10-01T00:00Z). Searched the PDF (all 8 pages), metadata/meta1-4.csv: nothing on whether a stamp is the start or end of its interval nor on the day definition. meta2.csv SCHEDULE column only lists the terms, e.g. ‘T,Teplota,°C,2.08,”AVG,06:00,13:00,20:00”’. Data checks (not documentation), Cheb Oct 2024: SRA (DT D 06:00Z) = sum of 10-min SRA10M over (D 06:00, D+1 06:00] UTC in 10/10 wet days (the preceding 24 h: 0/10); TMA/TMI (DT D 20:00Z) = 10-min extremes over (D-1 20:00, D 20:00] UTC (29/29, 24/29) better than the UTC day (27/29, 19/29); T AVG = (T06 + T13 + 2*T20)/4 (2025-12-31: (-7.1-2.1-0.8)/4 = -2.5).

  • observation/monthly: Library: Row picked by TIMEFUNCTION+MDFUNCTION; timestamp built as YEAR-MONTH-01 00:00 UTC from the YEAR/MONTH columns (parser.py), i.e. the library names the month by its first day. Request bounds: naive -> UTC (model/request.py); output cut inclusively to [start, end] (model/values.py). Source: Czech: “Měsíční klimatologické charakteristiky (průměr, úhrn, maximum, minimum, počet charakteristických dní) odvozeny z denních klimatologických charakteristik jsou dostupná v adresáři /monthly/” [monthly characteristics (mean, total, max, min, count of characteristic days) derived from the daily characteristics are in /monthly/] https://opendata.chmi.cz/meteorology/climate/Klimatologicka_data_popis.pdf. The file has YEAR,MONTH columns, no time stamp; a month is the set of its daily values, whose day definition is unverified (see daily).

  • observation/annual: Library: Row picked by TIMEFUNCTION+MDFUNCTION; timestamp built as YEAR-01-01 00:00 UTC (parser.py). Request bounds: naive -> UTC (model/request.py); output cut inclusively to [start, end] (model/values.py). Source: Czech: “Roční klimatologické charakteristiky (průměr, úhrn, maximum, minimum, počet charakteristických dní) jsou dostupná v adresáři /yearly/ v adresářích po charakteristikách” [annual characteristics … are in /yearly/] https://opendata.chmi.cz/meteorology/climate/Klimatologicka_data_popis.pdf. File has a YEAR column only; no time zone/period definition stated.

dmi#

network

resolution

datasets

source time zone

reference time

verified

observation

hourly

data

UTC

start (from = start of the hour)

yes

observation

daily

data

Europe/Copenhagen local time (CET/CEST) for Danish stations

start (local civil day, labelled with its date at 00:00 UTC)

yes

observation

daily

data

Europe/Copenhagen local time (CET/CEST)

end (precipitation_amount_max = max_precip_30m: window 23:31 local D-1 to 24:00 local D, ends on day D and is labelled D)

yes

observation

daily

data

UTC (Greenland/Faroe ‘old meteorological day’)

start (Greenland station values: 06:00 UTC D to 06:00 UTC D+1, labelled D)

yes

observation

monthly

data

Europe/Copenhagen local time (CET/CEST) for Danish stations

start (local calendar month, labelled day 1 00:00 UTC)

yes

observation

annual

data

Europe/Copenhagen local time (CET) for Danish stations

start (local calendar year, labelled Jan 1 00:00 UTC)

yes

Notes:

  • observation/hourly (data): Library: api.py parses ‘from’ with its offset and converts to UTC; request bounds astimezone(UTC) without margin (api.py); framework trims inclusively. Source: “Hourly station values are always in UTC. This avoids any confusion that might arise if summer- and winter time was used.” (https://www.dmi.dk/friedata/dokumentation/climate-data); “properties.from … formatted string with the start time of the time spanned by the station value”; “properties.to … formatted string with the end time of the time spanned by the station value” (https://www.dmi.dk/friedata/dokumentation/apis/climate-data-api-1). Observed: from 2025-03-30T01:00:00+00:00, to 2025-03-30T02:00:00+00:00 (station 06180).

  • observation/daily (data): Library: api.py takes the civil date from the first 10 characters of ‘to’ (the end of the day) minus one day and stamps 00:00 UTC of that date; query widened by +-1 day (api.py) and trimmed by the framework on the nominal stamp. Source: “Local day from/to midnight in local time with changes to summer and winter time” (https://www.dmi.dk/friedata/dokumentation/climate-data); “properties.from … formatted string with the start time of the time spanned by the station value”; “properties.to … formatted string with the end time of the time spanned by the station value” (https://www.dmi.dk/friedata/dokumentation/apis/climate-data-api-1). Observed: from 2025-03-30T00:00:00.001000+01:00, to 2025-03-31T00:00:00+02:00 (06180, mean_temp).

  • observation/daily (data): Library: api.py takes the date from ‘to’ (the end of the day, minus one day), not from ‘from’ = ‘D-1T23:31:00+01:00’, so day D’s value is stamped D 00:00 UTC. Source: “The exeption to the rule only used by the parameter of maximum precipation in 30 minute intervals. This is defined as the local day, but also includes the 30 minutes before midnight, so the maximum precipation at midnight is included.” and max_precip_30m “is found by inspecting the period from 23:31 local time to the previous day to 24.00 local time of the respective day” (https://www.dmi.dk/friedata/dokumentation/climate-data). Observed: from 2025-03-29T23:31:00+01:00, to 2025-03-31T00:00:00+02:00, value 3.8.

  • observation/daily (data): Library: api.py takes the date of ‘to’ (D+1) minus one day and stamps D 00:00 UTC, 6 h before the window starts. Source: “Old meteorological day a day which covers the hours from 6:00 am to 6:00 am next day in UTC”; “Time resolutions also differ between stations values for Denmark and Greenland/Faroe Islands” (https://www.dmi.dk/friedata/dokumentation/climate-data). Observed for Nuuk 04250: every daily parameter from 2025-03-30T06:00:00+00:00 to 2025-03-31T06:00:00+00:00.

  • observation/monthly (data): Library: api.py stamps 00:00 UTC on day 1 of the year-month in ‘from’; +-1 day query margin, framework trims. Source: “Monthly station values are derived from daily station values. They begin when the first day of the month begins, and end when the last day of the month ends.” “The monthly station value for the parameter mean_temp in March 2018 for a Danish station would represent the time range: 1/3/2018 00:01 CET - 31/3/2018 24:00 CEST.” (https://www.dmi.dk/friedata/dokumentation/climate-data)

  • observation/annual (data): Library: api.py stamps 00:00 UTC on 1 January of the year in ‘from’; +-1 day query margin, framework trims. Source: “Yearly station values are derived from daily station values. They begin when the first day in the year begins (January 1st), and end when the last day in the year ends (December 31st).” “…for a Danish station would represent the time range: 1/1/2018 00:01 CET - 31/12/2018 24:00 CET.” (https://www.dmi.dk/friedata/dokumentation/climate-data)

dwd#

network

resolution

datasets

source time zone

reference time

verified

observation

1_minute

all

UTC (recent/, now/); MEZ (UTC+1, no DST) before 2000 in historical/

end (of the 1-minute interval for rain[e]H3; Pluvio Ott delayed ~5 min)

yes

observation

5_minutes

all

UTC (recent/, now/); MEZ (UTC+1, no DST) before 2000 in historical/

end (of the previous 5 minutes)

yes

observation

10_minutes

temperature_air

UTC (recent/, now/); MEZ (UTC+1, no DST) before 2000 in historical/

instant (TT_10, TM5_10; RF/PP/TD not qualified)

yes

observation

10_minutes

precipitation, solar, wind, wind_extreme, temperature_extreme

UTC (recent/, now/); MEZ (UTC+1, no DST) before 2000 in historical/

end (of the previous 10 minutes)

yes

observation

10_minutes

urban_temperature_air

UTC

instant

yes

observation

10_minutes

urban_temperature_extreme, urban_wind_extreme, urban_precipitation, urban_pressure, urban_temperature_soil, urban_solar, urban_wind

unverified

unverified

no

observation

hourly

temperature_air

per station and period in Metadaten_Parameter: MEZ (UTC+1, no DST) where the period says ‘MEZ (MD)’ (80 East German stations, up to 1991), shifted to UTC by the library (confirmed against NOAA ISD at 38 of the 39 stations with a twin, 01048 in 1985 among them; the label alone for the others)

unverified (observed: instant, the value of the shifted stamp HH equals the minute-00 ISD report of HH UTC; the SYNOP 10 min offset of the UTC periods is not stated for these periods)

partly

observation

hourly

temperature_air

per station and period in Metadaten_Parameter: UTC (e.g. 01048 from 1991-11-19, 01975 from 1995-09-01); a plain ‘MEZ’ (UTC+1, no DST) period (160 stations, analog Thermohygrograph; observed: UTC at 28 stations, e.g. 01975 in 1990 to 1992, but MEZ at Diepholz from 1995-02-01 to 1996-05-31, which the library shifts, and at Regensburg-Oberhub and Stötten in 1983, which it does not: their periods are not settled)

mixed (UTC periods: instant, the SYNOP convention that the value of hour HH is the reading at HH-10 min UTC; analog MEZ periods: not stated)

partly

observation

hourly

precipitation

UTC (every period of all 1494 stations, per Metadaten_Parameter; e.g. 01048 from 1995-09-01)

unverified (observed: end (sum over (HH-2):50..(HH-1):50 UTC, i.e. the hour ending 10 min before the stamp))

partly

observation

hourly

solar

unverified

unverified (MESS_DATUM is a minute-resolution instant next to MESS_DATUM_WOZ ‘local true solar time’)

no

observation

hourly

wind

per station and period in Metadaten_Parameter: MEZ (368 of 453 stations carry MEZ periods) or UTC (observed: UTC also where the period says MEZ, in one or two years per station between 1973 and 2000 at about 45 stations, and at Bückeburg and Celle in 1990, where NOAA ISD is an hour off, not the DWD stamps)

mixed (periods generated from SYNOP reports, 1967 to 1991 in the eMD archive: the 10 minute mean at the full hour, ‘keine Stundenmittel’; periods from analog registrations, MIRIAM and AMDA: hourly mean, ‘Stundenmittel’, whose position in the hour is not stated)

partly

observation

hourly

cloud_type, cloudiness, dew_point, moisture, pressure, visibility, weather_phenomena, wind_extreme, wind_synoptic

UTC in the sampled Metadaten_Parameter files (every 7th station; the others were not read)

unverified

partly

observation

hourly

sun

per station and period in Metadaten_Parameter: WOZ (local true solar time) for the older periods (528 of 636 stations; most end on 2001-03-31, those of 172 stations run on, to 2006), UTC for the SYNOP-based periods (the WOZ stamps are returned unshifted and flagged zone:woz, #2816)

unverified (hourly sum; the window is not stated)

partly

observation

hourly

temperature_soil

per station and period in Metadaten_Parameter, by the source text (it has no zone column): MEZ (UTC+1, no DST) where the source is the eMD archive (74 East German stations, up to 1991: ‘01,07,13,19 MEZ und synoptische Haupttermine 01,04,…,22 MEZ’; observed: stamps 01,04,…,22 only, a diurnal cycle about 1.4 h ahead of the SYNOP periods), shifted to UTC by the library; the later climate terms (‘07, 14, 21 MOZ, ab 01.01.1987 um 07:30,14:30,21:30 MEZ’) and the SYNOP periods (24 UTC hours) as written

unverified

partly

observation

hourly

urban_temperature_air

UTC

instant (reading 10 min before the stamp)

yes

observation

hourly

urban_precipitation

UTC

end (of the preceding hour)

yes

observation

hourly

urban_pressure, urban_sun, urban_temperature_soil, urban_wind

unverified

unverified

no

observation

subdaily

temperature_air

per station and period in Metadaten_Parameter (UTC, MEZ, MOZ); against NOAA ISD the UTC periods and the periods labelled MEZ from 1973 to 2000 are UTC in the values (the stamp is the UTC hour of the term), the MEZ label up to 1946 and the MOZ label up to 1972 look real (lag -1 or -2) in a minority of station-years, MOZ being local mean time and no whole hour

instant (observation term, ‘Terminwert’)

partly

observation

subdaily

cloudiness, moisture, pressure, soil, visibility, wind, wind_extreme

unverified

unverified

no

observation

daily

climate_summary

mixed per parameter and period (Metadaten_Parameter): UTC since 2001-04-01 at 01048; MEZ/MOZ windows 1991-2001 and before 1945; UTC windows 1960-1990 (GDR)

mixed (date label; e.g. RSK 05:50 UTC D -> 05:50 UTC D+1, TXK/TNK 00-24 UTC, SHK_TAG 06 UTC instant)

partly

observation

daily

precipitation_more, solar, temperature_soil, water_equivalent, weather_phenomena, weather_phenomena_more

unverified

unverified

no

observation

monthly

climate_summary

unverified (aggregates of daily values; inherits the daily windows)

start (MESS_DATUM_BEGINN, first day of the month)

partly

observation

monthly

climate_indices, precipitation_indices, precipitation_more, weather_phenomena

unverified

unverified

no

observation

annual

all

unverified

start (MESS_DATUM_BEGINN, 1 January)

partly

mosmix

hourly

all

UTC

instant (forecast point in time: TTT, Td, T5cm, DD, FF, PPPP, N*, VV, ww …)

partly

mosmix

hourly

all

UTC

end (of the preceding 1/3/6/12/24 h: RR1c, RR3c, RR6c, RRhc, RRdc, FX1, FX3, FXh, TX, TN, TG, TM, SunD1, SunD3, RSunD, PEvap, R*/ww* probabilities, W1W2, Rad1h)

partly

mosmix

hourly

large

UTC

unverified (SunD: ‘yesterday’, carried at the 06 UTC step; which day boundary is not stated)

partly

dmo

hourly

all

UTC

instant (ttt, td, t5cm, dd, ff, pppp, n, nh, nl, nm, neff, ww)

partly

dmo

hourly

all

UTC

end (of the preceding 1/3/6/12 h: rr1, rr3, rrs1c, rrs3c, fx3, rads3, radl3, tx, tn, w1w2)

partly

road

15_minutes

all

unverified

instant (airTemperature, dewpointTemperature, relativeHumidity, horizontalVisibility, roadSurfaceTemperature/-Condition, waterFilmThickness, precipitationType/intensity)

partly

road

15_minutes

all

unverified

end (of the preceding 15 min: totalPrecipitationOrTotalWaterEquivalent)

partly

road

15_minutes

all

unverified

unverified (windSpeed/windDirection documented as 10-min mean ending at the stamp, but the real file states 0 min; maximumWindGust* period ‘In minutes’, missing in the real file)

no

swsmos

hourly

all

UTC

instant (TL, TD, TS, RC at the forecast time)

yes

swsmos

hourly

all

UTC

end (of the preceding 1 h: RRL1c; 6 h: RR6, WWL6, R650)

yes

poi

hourly

all

UTC

mixed (instant at the report time for temperature, dew point, humidity, pressure, cloud, visibility, snow depth, present_weather; end for wind: the mean of the last 10 minutes before the stamp)

partly

poi

hourly

all

UTC

end (of the preceding 1/3/6/12/24 h: precipitation_amount_last_*, maximum/minimum_temperature_last_12_hours, maximum_wind_speed_last_hour/6_hours, maximum_of_10_minutes_mean_of_wind_speed_for_previous_day (wind_speed_rolling_mean_max_last_6h, carried at 00, 06, 12 and 18 UTC), *_radiation_last_hour, total_time_of_sunshine_during_last_hour, evapotranspiration_last_24_hours, past_weather_1/2 at 3-hourly synoptic hours)

yes

poi

hourly

all

UTC

unverified (previous-day values, carried only in the 06:00 UTC report, except the wind maximum, which is a 6-hour maximum, see the row above; day boundary not stated)

partly

phenology

annual

all

unverified

unverified (observed: start (of the Referenzjahr, a library choice; the phase itself is a calendar day, Eintrittsdatum, which is dropped; value = Jultag))

no

radar

n/a

all

UTC

mixed (measurement time; accumulations over INT minutes: window position relative to the stamp not stated in the quoted text; RADVOR: stamp = forecast base time, valid time = stamp + VV minutes)

partly

alerts

n/a

all

Europe/Berlin local time (CET/CEST) with explicit UTC offset in every stamp

instant (sent/effective = issue time, onset = start, expires = end of the event)

yes

derived

hourly

radiation_global

UTC

end (of the synoptic hour, window (HH-2):50..(HH-1):50 UTC, i.e. ending 10 min before the stamp)

yes

derived

hourly

sunshine_duration

unverified (UTC stated for radiation_global only)

end (of the synoptic hour, window (HH-2):50..(HH-1):50)

partly

derived

daily

soil

unverified

unverified

no

derived

monthly

soil

unverified

unverified

no

derived

monthly

heating_degreedays, cooling_degreehours_13, cooling_degreehours_16, cooling_degreehours_18

unverified

start (month named by the file, ‘Monat yyyymm’)

partly

derived

monthly

climate_correction_factor

unverified

start (of the 12-month window KF_<begin>_<end>)

partly

Notes:

  • observation/1_minute: Library: historical: MESS_DATUM_BEGINN/ENDE parsed as UTC and exploded to one row per minute (src/wetterdienst/provider/dwd/observation/parser.py); recent/now: MESS_DATUM parsed as UTC (parser.py); then -1 h for the historical files whose name begins before 2000, as a whole (parser.py). Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: “The measurements are assigned to a time stamp in MEZ before the year 2000, and to a time stamp in UTC from the year 2000 dedicated.” (historical/) and “The time stamp is given in UTC.” (recent/, now/); “The meaning of the time stamp is instrument dependent. For the instrument rain[e]H3, the time stamp denotes the end of the measurement inverval. In case of instrument Pluvio Ott a time delay of approximately 5 minutes is introduced due to the inertia of the instrument response.” – https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/1_minute/precipitation/DESCRIPTION_obsgermany-climate-1min-rr_en.pdf

  • observation/5_minutes: Library: Parsed ‘%Y%m%d%H%M’ as UTC (src/wetterdienst/provider/dwd/observation/parser.py); then the parser subtracts 1 h from every stamp of a historical file whose name begins before 2000 (parser.py, _mez_to_utc), i.e. reads the files up to 1999 as MEZ = UTC+1 without DST and those from 2000 as UTC, whatever the year of a stamp in them (the December 1999 file holds a row stamped 200001010000, which is not a reading of that file and is dropped: it is the 00:00 UTC row made of the 23:56 to 23:59 UTC minutes the file has, checked against the 1-minute precipitation of 136 stations, and the January file holds it whole; the file has no 199912312355 row, and the 5-minute data of 22:55 to 23:55 UTC does not exist). Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: “The measurements are assigned to a time stamp in MEZ before the year 2000, and to a time stamp in UTC from the year 2000 dedicated.” (historical/) and “The time stamp is given in UTC.” (recent/, now/); “RS_05 [\Sum of the precipitation height of the previous 5 minutes”; “For the instrument rain[e]H3, the time stamp denotes the end of the measurement inverval.” – https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/5_minutes/precipitation/DESCRIPTION_obsgermany-climate-5min-rr_en.pdf

  • observation/10_minutes (temperature_air): Library: Parsed ‘%Y%m%d%H%M’ as UTC (src/wetterdienst/provider/dwd/observation/parser.py); then the parser subtracts 1 h from every stamp of a historical file whose name begins before 2000 (parser.py, _mez_to_utc), i.e. reads the files up to 1999 as MEZ = UTC+1 without DST and those from 2000 as UTC, whatever the year of a stamp in them (the 10-minute wind file of 1999 holds MEZ stamps up to 2000-01-01 00:50). Checked at 01048: after the correction, NOAA ISD SYNOP of hour HH in 1992 equals library tt_10 at HH-20min in 80% of 7319 hours, and library hourly-vs-10-minute offsets are the same in 1999 (corrected) and 2000 (UTC, uncorrected), so no 1 h step at the switch. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: “The measurements are assigned to a time stamp in MEZ before the year 2000, and to a time stamp in UTC from the year 2000 dedicated.” (historical/) and “The time stamp is given in UTC.” (recent/, now/); “TT_10 [\air temperature 2 m above ground\,\instant\,\missing_value=-999\]”; de: “Die Messungen sind vor dem Jahr 2000 einem Zeitstempel in MEZ und ab dem Jahr 2000 einem Zeitstempel in UTC zugeordnet.” – https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/10_minutes/air_temperature/DESCRIPTION_obsgermany_climate_10min_air_temperature_en.pdf and …/BESCHREIBUNG_obsgermany_climate_10min_air_temperature_de.pdf

  • observation/10_minutes (precipitation, solar, wind, wind_extreme, temperature_extreme): Library: Parsed ‘%Y%m%d%H%M’ as UTC (src/wetterdienst/provider/dwd/observation/parser.py); then the parser subtracts 1 h from every stamp of a historical file whose name begins before 2000 (parser.py, _mez_to_utc), i.e. reads the files up to 1999 as MEZ = UTC+1 without DST and those from 2000 as UTC, whatever the year of a stamp in them (the 10-minute wind file of 1999 holds MEZ stamps up to 2000-01-01 00:50). The file of 1999 of precipitation, extreme_temperature and extreme_wind lists each of 199912312300 to 199912312350 twice: one row is the reading at 23:xx MEZ, the other a copy of the reading at 00:xx MEZ of 2000-01-01, which the file of 2000 holds as 199912312300 to 199912312350 UTC (of 226 stamps with two different rows, exactly one equals the row of the file of 2000; the 1-minute rain at 00:19, 00:27 and 00:46 MEZ is in the copy of the stamps 00:20, 00:30 and 00:50). The parser keeps the row that is not a copy (parser.py, _drop_mez_copies_of_utc_rows), which is why a request touching 22:00 UTC on that day also reads the file of 2000. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: “The measurements are assigned to a time stamp in MEZ before the year 2000, and to a time stamp in UTC from the year 2000 dedicated.” (historical/) and “The time stamp is given in UTC.” (recent/, now/) (same sentence in each); precipitation: “RWS_10 [\Sum of the precipitation height of the previous 10 minutes” + “For the instrument rain[e]H3, the time stamp denotes the end of the measurement inverval”; solar: “GS_10 [\sum of global radiation during the previous 10 minutes”; wind: “FF_10 [\mean wind speed during the previous 10 minutes”; extreme_wind: “FX_10 [\maximum of wind speed during the previous 10 minutes”; extreme_temperature: “TX_10 [\maximum of air temperature at 2 m height during the last 10 minutes” – https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/10_minutes/{precipitation/DESCRIPTION_obsgermany-climate-10min-precipitation_en.pdf, solar/DESCRIPTION_obsgermany_climate_10min_solar_en.pdf, wind/DESCRIPTION_obsgermany_climate_10min_wind_en.pdf, extreme_wind/DESCRIPTION_obsgermany_climate_10min_extreme_wind_en.pdf, extreme_temperature/DESCRIPTION_obsgermany_climate_10min_extreme_temperature_en.pdf}

  • observation/10_minutes (urban_temperature_air): Library: Parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py); no shift (urban data start 2015). Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: Metadaten_Parameter_zehn_st_min_tu_00399.txt: “TT_ST_10;momentane Lufttemperatur in 2m Hoehe;…;in UTC” (de: momentary air temperature, in UTC) – inside https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate_urban/10_minutes/air_temperature/recent/10minutenwerte_TU_STADT_00399_akt.zip (no description PDF in that directory)

  • observation/10_minutes (urban_temperature_extreme, urban_wind_extreme, urban_precipitation, urban_pressure, urban_temperature_soil, urban_solar, urban_wind): Library: Parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py), no shift (urban data start 2015). Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: Not read: only the urban 10-minute air_temperature metadata was fetched (throttling); climate_urban/10_minutes/air_temperature/recent/ has no description PDF.

  • observation/hourly (temperature_air): Library: MESS_DATUM ‘YYYYMMDDHH’ padded to ‘%Y%m%d%H%M’ and parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py). For a period that Metadaten_Parameter calls ‘Stundenwerte in MEZ (MD)’ (the eMD archive of the DDR’s meteorological service: 80 of the 637 historical zips, up to 1991) the stamp is MEZ, the fixed UTC+1, and is shifted by -1 h to UTC; the rows carry zone:mez in quality_flags (GH-2709, download.py _read_mez_md_spans, parser.py, api.py _tidy_up_df). The shift is made per parameter, as TT_TU and RF_TU share one MESS_DATUM column but have different period tables in 42 of 637 zips (Aue 00222: MD from 1982 for TT_TU, from 1991 for RF_TU), and only for this label: the plain ‘MEZ’ and the ‘UTC’ periods, and a zip without Metadaten_Parameter, are returned as written. The periods are dates, and in 38 cases (TT_TU) an MD period ends on the day the next period begins (36 of them on 1991-11-01, Fichtelberg 1991-05-02, Berlin Brandenburg 1991-07-01; three MD periods begin on a day the period before ends: Fichtelberg 1991-05-11, Harzgerode 1991-10-02, Berlin-Alexanderplatz 1975-04-01), so the clock changes inside that day at an hour the metadata does not give. Against NOAA ISD (minute-00 reports, 2026-10-16) the change is after the stamp 00 of that day: the stamp 00 is still MEZ in 18 of 18 cases (it is the 23 UTC report of the day before), the stamps 01 to 03 are UTC in 11 of 11 readings at 10 stations (Fichtelberg 1991-05-02, Chemnitz, Leinefelde and Oschatz on 1991-11-01 among them), and 75 of the 88 later stamps of the day agree with UTC, 13 with MEZ (Ummendorf 05158 reads MEZ at the stamps 09 to 12 of 1991-11-01 and again on 1991-11-04). Treating the shared day as MEZ throughout would mislabel 21 to 23 stamps per case. The rule is therefore: the stamp 00 of a shared day belongs to the period that ends on it, the stamps 01 to 23 to the period that begins on it, so an MD period that ends on a shared day ends at its stamp 00 and one that begins on it starts at its stamp 01; any other MD period runs from the first stamp of its first day to the last of its last day. Where the shift puts a row on a stamp that is there already (30 stamps at 26 stations, at the beginning of an MD period; 28 hold the same value, null included, on both) the row that was not moved is kept. Checked against NOAA ISD (UTC, one year inside the period per station, the reader’s output at the same UTC hour, 2026-10-16): 95.3 to 99.9 % of the values are identical in tenths of a degree at 38 of the 39 stations with a twin that reports in an MD period (5.0 to 15.7 % before the shift), e.g. 01048 (ISD 09488) 1985 96.0 % of 8727 hours (8.4 % before) and Fichtelberg 1991 99.7 % (12.2 % before); Cottbus is the exception at 50.8 % (6.6 % before); no station contradicts, and the other stations follow from the label alone. The reference time of these periods is not stated; the SYNOP sentence quoted in the next note is for the UTC periods. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: Metadaten_Parameter_tu_stunde_01048.txt: “1048;19730101;19911101;Dresden-Klotzsche;TT_TU;Lufttemperatur;°C;Daten der Temperatur und relativen Feuchte (Stundenwerte) aus dem Archiv des eMD (EMDS) bis 1991;Stundenwerte in MEZ (MD)” and “1048;20010401;20260418;…;TT_TU;…generiert aus SYNOP-Meldungen (nach dem 01.04.2001);Stundenwerte in UTC” (de: hourly values in MEZ / in UTC) – inside https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/hourly/air_temperature/historical/stundenwerte_TU_01048_19730101_20251231_hist.zip. The dataset PDF https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/hourly/air_temperature/DESCRIPTION_obsgermany_climate_hourly_air_temperature_en.pdf (and _de) states no time zone.

  • observation/hourly (temperature_air): Library: Parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py), no shift, for the UTC periods and for the plain ‘MEZ’ periods. The plain ‘MEZ’ label does not predict a shift. Against NOAA ISD (UTC): Hamburg-Fuhlsbüttel 01975 (ISD 10147) 1990 to 1992, labelled MEZ until 1995-08-31, equals ISD at lag 0 in 99.2 % of 25679 hours and at lag -1 in 6.8 %, so its values are UTC; 27 more stations with a MEZ period match at lag 0 with 97 % or more (Aachen, Altenstadt, Bad Marienberg, Bamberg, Bonn-Friesdorf, Bremerhaven, Cuxhaven, Deuselbach, Emden, Hannover, Ingolstadt-Manching, Karlsruhe, Kassel, Köln-Bonn, Lingen, Michelstadt, Münster-Osnabrück, Nürburg, Oldenburg, Osnabrück-Atterheide, Straubing, Stuttgart-Echterdingen, Tegel, Tholey, Trier-Petrisberg, Wasserkuppe, Zugspitze). Three match at lag -1 instead (Diepholz 1995 92.0 %, Regensburg-Oberhub 1983 87.1 %, Stötten 1983 85.3 %; the library shifts Diepholz and not the other two, below). Day by day against ISD (minute-00 reports, read 2026-10-17; a day is told apart when its mean absolute difference at one lag is under 0.25 K and at the other more than twice that plus 0.2 K) the stamps change clock inside the labelled periods, not where the label does. Diepholz 00963 (ISD 103210; label MEZ 1994-04-01 to 1996-06-01) is UTC up to the stamps of 1995-01-31 (24 of 24 hours identical at lag 0 on that day, 99 % in 1994) and MEZ from the stamp 1995-02-01 00 to 1996-05-31 (24 of 24 hours identical at lag -1 on both days, 99 to 100 % in each month between), then UTC again from 1996-06-01. A source that does not depend on ISD agrees (read 2026-10-23): the first differences of the hourly values against the mean of the 8 nearest stations within 130 km that DWD labels UTC for the whole time (Bremen, Hannover and Soltau among them) correlate best at lag 0 in each month from 1994-04 to 1995-01 and in 1996-06 and 1996-07, and at lag -1 in each of the 16 months from 1995-02 to 1996-05 (relative humidity: lag -1 beats lag 0 in 14 of the 16 months, lag 0 before and after). Over the whole MEZ span 11,374 of 11,397 hours equal ISD at lag -1 against 1,075 at lag 0; before it 729 of 735 equal it at lag 0 against 68 at lag -1, after it (to 1996-12-31) 4,668 of 4,745 against 345. The library therefore shifts the stamps of Diepholz from 1995-02-01 00 to 1996-05-31 23 by -1 h, both parameters, and flags them zone:mez (MEZ_BY_COMPARISON in download.py, only inside the period the metadata still calls plain MEZ); the stamp 1995-02-01 00 is the report of the stamp 1995-01-31 23, which the file has, and its row is dropped. Regensburg 04104 (ISD 107760; label MEZ 1948-01-01 to 1994-03-31, and MEZ in alternate months of 1998 to 2004): every day told apart from 1980-01-13 to 1985-12-31 is MEZ, from 1986-01-03 to 1986-05-31 UTC, from 1986-06-01 to 1987-11-26 MEZ, from 1987-12-01 to 1988-02-12 UTC, from 1988-02-13 to 1988-03-27 mixed (MEZ, UTC, MEZ) and from 1988-03-28 to the end of ISD in 1992 UTC; before 1980 the days are not told apart, and ISD has no 1993 or 1994. Stötten 04887 (ISD 108360; label MEZ 1948 to 1994-12-31): every day told apart from 1979-09-04 to 1985-04-14 is MEZ, 1985-04-15 to 1985-07-15 mixed, and 1985-07-16 to the end of ISD in 1992 UTC. The periods that are MEZ in the values can therefore be read off neither the label nor a year per station; the 160 stations with a plain MEZ period were not scanned this way, and the library returns every one of them as written, Regensburg-Oberhub and Stötten among them: the second source does not settle their periods (#3029). Against the mean of the 8 nearest stations labelled UTC or plain MEZ (2026-10-23) Regensburg is at lag -1 in 1986-06 to 1987-11 and at lag 0 in 1987-12 to 1988-01 and from 1988-04 (August aside), as ISD has it, but at lag 0 in 1985-04 to 1985-08 where ISD has MEZ, and Stötten is at lag 0 in 1983 and 1984 and at lag +1 from 1985-04 to 1986-12, which fits ISD (MEZ to 1985-04-14, UTC from 1985-07-16) only if those neighbours are MEZ all the time; they are plain MEZ stations whose own clock is not known, ISD ends in 1992 (none for 1993 and 1994), and before 1980 the days are not told apart, so no list of stamps is made for them. In many other station-years (about 40 stations in at least one year) no ISD series agrees with the values at any lag (a different observation system), so those periods are unverified. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: Metadaten_Parameter_tu_stunde_01048.txt: “1048;20010401;20260418;…;TT_TU;…generiert aus SYNOP-Meldungen (nach dem 01.04.2001);Stundenwerte in UTC” (de: hourly values in UTC) and Metadaten_Parameter_tu_stunde_01975.txt: “1975;19490101;19950831;Hamburg-Fuhlsbüttel;TT_TU;…(Thermohygrograph);Stundenwerte in MEZ” and “1975;19950901;20010401;…;Stundenwerte in UTC” – inside https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/hourly/air_temperature/historical/stundenwerte_TU_01975_19490101_20251231_hist.zip; reference time from the urban sibling: “Hourly air temperature and humidity are given following the SYNOP conventions. The hour x corresponds to the measurement from 10min before, (e.g., at UTC 11, the mean value of the 1 minute interval ending UTC 10:50 is given).” – https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate_urban/hourly/air_temperature/recent/DESCRIPTION_obsgermany_climate_urban_hourly_tu_recent_en.pdf. The dataset PDF states no time zone. Data: 2025-04..2026-10 TT_TU(HH) == TT_10(HH-10min) in 100% of 13197 hours.

  • observation/hourly (precipitation): Library: Parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py), no shift. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: every period of the Metadaten_Parameter files of all 1494 historical zips reads ‘in UTC’ (or ‘01,02,04,…,23 UTC=1h Bezugszeitraum’); Metadaten_Parameter_rr_stunde_01048.txt: “19950901;20010331;R1;stdl. Niederschlagshoehe;in UTC” and “20010401;20260419;R1;stdl. Niederschlagshoehe;in UTC” – inside https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/hourly/precipitation/historical/stundenwerte_RR_01048_19950901_20251231_hist.zip. Reference time not stated in the documents read; data: R1(HH) equals the sum of RWS_10 over the 60 min ending HH-10min (MAE 0.032 mm, 77% identical) vs ending HH (MAE 0.168).

  • observation/hourly (solar): Library: MESS_DATUM ‘YYYYMMDDHH:MM’ parsed as UTC and rounded to the nearest hour (src/wetterdienst/provider/dwd/observation/parser.py); true_local_time_offset kept as minutes between the two stamps (parser.py). Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: “MESS_DATUM reference date VARCHAR2 YYYYMMDDHH24:mm”; “MESS_DATUM_WOZ local true solar time”; “ZENIT [\solar zenith angle at mid of interval” – https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/hourly/solar/DESCRIPTION_obsgermany_climate_hourly_solar_en.pdf (no time zone stated).

  • observation/hourly (wind): Library: Parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py), no correction for MEZ periods. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: 368 of the 453 stations have a MEZ period in their Metadaten_Parameter file, the newer periods read UTC, e.g. Artern: “198;19911107;20030831;Artern;F;Windgeschwindigkeit Messnetz 3;m/sec;Winddaten (Stundenmittel, maximale Windspitze 00:00-23:59 MEZ) generiert aus 10-Minutenmittel von automatischen Stationen der 1.Generation (MIRIAM/AFMS2), Richtungsangaben in 36-teiliger Windrose;Stundenwerte in MEZ” and “198;20030901;20260418;…;Winddaten (Stundenmittel, maximale Windspitze 23:51-23:50 UTC) generiert aus 10-Minutenmittel von automatischen Stationen der 2. Generation (AMDA), Richtungsangaben in 36-teiliger Windrose;Stundenwerte in UTC” – inside https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/hourly/wind/historical/stundenwerte_FF_00198_19610101_20251231_hist.zip. Observed against NOAA ISD (UTC), correlation of F with the ISD wind speed by lag in one or two years per station: highest at lag 0 almost everywhere, also inside MEZ periods, e.g. at Brocken 0.999, Erfurt-Bindersleben 0.998, Artern 0.997 and Berlin-Schönefeld 0.996 (East Germany, 1983 and 1991) and Bremen 0.978; only Bückeburg 00769 and Celle 00850 peak at lag -1, in 1979 to 1981 and 1988 to 1990 (1990: 0.971 against 0.927 at lag 0 for Bückeburg, 0.971 against 0.929 for Celle, about 1900 hours each; 1979: 0.975 against 0.914 and 0.976 against 0.909), and at lag 0 from 1993 (and in 1982 to 1987 and 1991 to 1992 on 50 to 140 hours a year). It is NOAA ISD that is an hour off there, not the DWD stamps (read 2026-10-23, GH-3029): the hourly F of Celle correlates best with that of DWD Hannover 02014 at lag 0 in 1980, 1985, 1989, 1990 and 1995 alike (1990: 0.860 at lag -1, 0.871 at 0, 0.848 at +1; the same with Braunschweig 00662 and Soltau 04745), so the DWD series is not shifted in 1990 against its neighbours, and the peak of Bückeburg against the same four neighbours sits at lag 0 to +1 in 1980, 1985, 1989, 1990 and 1995 alike; ISD Celle and ISD Bückeburg against ISD Hannover peak at lag +1 in 1989 and 1990 (Celle 1990: 0.818 at +1 against 0.809 at 0; Bückeburg 1989: 0.775 against 0.749), while DWD Celle against ISD Hannover peaks at lag 0 (0.846 against 0.836 at -1). The MEZ label is therefore not applied. Reference time: the Artern eMD periods read “Winddaten (Stundenwerte als 10-Minutenmittel) generiert aus SYNOP-Meldungen” and the F remark “Im Zeitraum 01.04.1972 bis 31.10.1991 wurden die Werte der Windgeschwindigkeiten zur Stunde (keine Stundenmittel) in ganzen m/s erhoben”; the periods generated from analog registrations (e.g. Artern 1961 to 1966, Bremen to 1991) and from the MIRIAM and AMDA automatic stations read “Stundenmittel”, with no position in the hour. The dataset description gives no zone.

  • observation/hourly (cloud_type, cloudiness, dew_point, moisture, pressure, visibility, weather_phenomena, wind_extreme, wind_synoptic): Library: Parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py), no MEZ period was found in the sample, so no correction is applied. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: the Metadaten_Parameter files of the zips of every 7th station of each of these datasets (44 to 103 files per dataset) name UTC for every period and none names MEZ – e.g. https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/hourly/moisture/historical/. The dataset PDFs were not read for a reference time, except wind_synop’s. Wind_synoptic (wind_speed_mean_last_10min = FF, wind_direction = DD): Metadaten_Parameter_f_stunde_00011.txt “FF;10-Min-Mittel der Windgeschwindigkeit;m/sec;Daten aus SYNOP-Meldungen …;in UTC” (de: 10-minute mean of the wind speed, data from SYNOP reports); neither it nor the dataset PDF (“time series, hourly value”) says where the 10 minutes sit relative to the stamp, so that is unverified.

  • observation/hourly (sun): Library: Parsed as written and labelled UTC (src/wetterdienst/provider/dwd/observation/parser.py); a WOZ period is not shifted, as the true solar hour of a station is no whole hour from UTC (longitude over 15 plus the equation of time: about +35 min at Aachen, +20 at Hamburg, 0 at Görlitz, plus or minus 16 min over the year), but its rows carry zone:woz in quality_flags (#2816). The periods are read per station from the Metadaten_Parameter file of the same archive (download.py), not taken from one date: of the 636 historical archives (read 2026-10-15) 140 of the 528 stations with a WOZ period have their last one end on 2001-03-31, 172 have one that runs past it, to 2006 (249,640 station days), and some change between WOZ and UTC by the month (00769 in 2004). The deviation is confirmed against the stamps of the sunshine profile, not against a second source of the duration: the midpoint of the sunshine of a June day is 12.0 to 12.15 at six stations from 6.1 E to 15.0 E in the WOZ periods (true solar noon) and follows the longitude gradient (12.15 to 12.93) in the UTC ones. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: Metadaten_Parameter_sd_stunde_00198.txt: “198;19580401;19921130;Artern;SD_SO;stdl. Sonnenscheindauer;;Daten der Sonnenscheindauer (Stundensummen) aus analogen Registrierungen (Campbell-Stokes);Stundenwerte in WOZ” and “198;20010401;20220122;…;generiert aus stündlichen SYNOP-Meldungen (nach 01.04.2001);Stundenwerte in UTC” – inside https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/hourly/sun/historical/stundenwerte_SD_00198_19580401_20220122_hist.zip. 528 of the 636 stations have a WOZ period, 328 a UTC one. The dataset’s description PDF was not read.

  • observation/hourly (temperature_soil): Library: Parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py). For a period whose source Metadaten_Parameter gives as the archive of the eMD (“Daten der Erdbodentemperatur aus dem Archiv des eMD”: 408 period lines, 74 East German stations, up to 1991) the stamp is MEZ, the fixed UTC+1, and is shifted by -1 h to UTC, per parameter (depth), the rows carrying zone:mez in quality_flags (GH-3029, download.py _read_mez_md_spans, parser.py, api.py _tidy_up_df). The rule is the source text, as the Zusatz-Info of these periods is only Stundenwerte; the other periods (the climate terms, the SYNOP bulletin, the MIRIAM automatic stations, the Potsdam secular station) are returned as written. Evidence (read 2026-10-23, all 491 historical archives): (1) the description says the terms are MEZ (‘3,4 Terminwerte:01,07,13,19 MEZ und synoptische Haupttermine:01,04,07,10,13,16,19,22 MEZ’); (2) in the eMD periods of all 74 stations (605,757 station days) the stamps are 01, 04, 07, 10, 13, 16, 19 and 22 only, never 00, 03, …, 21, which are the synoptic main terms in UTC; the SYNOP periods of the same stations (from 2001-04-01) have all 24 hours and the subdaily files of the same stations hold the terms of those years at 00, 06, 12 and 18 UTC (01048: soil state EK_TER, and the air temperature, which equals NOAA ISD at the same UTC hour, #2711); (3) the diurnal cycle of the soil temperature: for the 56 stations with an eMD and a SYNOP period, the time of the maximum of the 24 h harmonic (days of June to August with the four terms 01, 07, 13, 19, the mean of the four removes the harmonic’s offset exactly) behind the maximum of the harmonic of the hourly air temperature of the same days (all 24 hours, UTC after the MEZ (MD) shift) is, with the stamps as written, 1.4 h larger in the eMD periods than in the SYNOP periods (UTC) at 5 cm (median of the stations; quartiles 1.1 and 1.7 h; 53 of 56 above 0.5 h, none negative), 1.2 h at 10 cm and 1.1 h at 20 cm; a stamp that is UTC would give 0, MEZ gives 1. The excess over 1 h at 5 cm is not explained (a different sensor and exposure in the eMD years is likely). No second source with soil temperature at the terms was found: NOAA ISD has none. The later period (1991-11-02 to 2004-12-31, ‘Klimaterminen’) has stamps at 07, 14 and 21 (01048 in 1995) for terms the description gives as 07:30, 14:30 and 21:30 MEZ, which would be 06:30, 13:30 and 20:30 UTC; that was not compared and the stamps are returned as written (unverified). The Potsdam secular station 03988 (‘4 Termine: 01, 07, 13, 19 MOZ’, 1949 to 2019, stamps 01, 07, 13, 19) is returned as written too: its diurnal cycle (5 cm) is 1.8 to 1.9 h behind the air temperature in 1976 to 1990 and in 2001 to 2019 alike, where the eMD periods of the other stations show 1.9 h (median) with the stamps as written and their SYNOP periods 0.6 h, so it sits where the eMD stamps do, but MOZ is local mean time at 13.1 degrees east (UTC+0.9 h) and no whole hour, so it was not shifted. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: Metadaten_Parameter_eb_stunde_01048.txt: “1048;19760409;19911031;Dresden-Klotzsche;V_TE005;Temperatur im Erdboden in 5 cm Tiefe;°C;Daten der Erdbodentemperatur aus dem Archiv des eMD (EMDS) von 1951-1991 (3,4 Terminwerte:01,07,13,19 MEZ und synoptische Haupttermine:01,04,07,10,13,16,19,22 MEZ);Stundenwerte” and “1048;19911102;20041231;…;Erdbodentemperaturen an den drei Klimaterminen ( 07, 14, 21 MOZ, ab 01.01.1987 um 07:30,14:30,21:30 MEZ) ;Stundenwerte” – inside https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/hourly/soil_temperature/historical/stundenwerte_EB_01048_19760409_20251231_hist.zip.

  • observation/hourly (urban_temperature_air): Library: Parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py), no shift. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: “Hourly air temperature and humidity are given following the SYNOP conventions. The hour x corresponds to the measurement from 10min before, (e.g., at UTC 11, the mean value of the 1 minute interval ending UTC 10:50 is given).” – https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate_urban/hourly/air_temperature/recent/DESCRIPTION_obsgermany_climate_urban_hourly_tu_recent_en.pdf

  • observation/hourly (urban_precipitation): Library: Parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py), no shift. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: “Hourly precipitation is given as sum of the six 10min intervals measured in the previous hour (e.g., at UTC 11, the sum of precipitation during UTC10-UTC11 is given).” – https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate_urban/hourly/precipitation/recent/DESCRIPTION_obsgermany_climate_urban_hourly_precipitation_recent_en.pdf

  • observation/hourly (urban_pressure, urban_sun, urban_temperature_soil, urban_wind): Library: Parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py), no shift. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: Their DESCRIPTION PDFs (climate_urban/hourly/<dataset>/recent/) were not fetched (throttling).

  • observation/subdaily (temperature_air): Library: MESS_DATUM ‘YYYYMMDDHH’ parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py); no correction for MEZ/MOZ periods, which is right where the label is wrong, and unproven where it is real (below). Data check (#2711, 2026-10-16, TT_TER and RF_TER of the subdaily zips against NOAA ISD, the key-less S3 bucket noaa-global-hourly-pds, minute-00 reports, nearest ISD station within 4 km; the DWD stamp HH against ISD at HH+lag; a station-year is determinate at the lag where at least 85 % of the values are within 0.5 K, and 10 points more than at any other lag; a random sample of up to 10 station-years per label and era): UTC 1947-1972 6 of 6 and 1973-1990 10 of 10 at lag 0, UTC 2001+ 9 of 10; MEZ 1991-2000 9 of 10 at lag 0 (the 07/14/21 terms of 01048 for 1991-2001, labelled MEZ there, equal ISD at 07/14/21 UTC, MAE 0.02 K; the 00/06/12/18 terms of 1985, labelled UTC, equal ISD at the same UTC hour, MAE 0.004 K), MEZ 1973-1990 4 of 10 at lag 0 and none at lag -1, MEZ 1947-1972 none determinate (best lag 0 in 7 of 10), MEZ 2001+ 1 of 7 at lag 0, MEZ up to 1946 3 of 8 at lag -1 (5 indeterminate); MOZ 1947-1972 4 of 10 at lag -1 (the other 6 have lag -1 as their best), MOZ up to 1946 2 of 10 at lag -1 and 2 at lag -2, MOZ 1991-2000 (the East German periods from 1991-11-01) 1 of 2 at lag 0 (Carlsfeld 1992: 100 % within 0.5 K). Relative humidity (RF_TER from ISD T and dew point, within 3 %, 8 station-years per cell): MEZ 1973-2000 10 of 16 determinate at lag 0 and none at another lag, MOZ 1947-1972 8 of 8 with best lag -1 but none determinate, UTC 2001+ 6 of 8 at lag 0. So the plain MEZ label of 1973 to 2000 is no whole-hour shift (a correction keyed on it would move values that are UTC already), and labels flip inside a station; the MEZ label up to 1946 and the MOZ label up to 1972 look real but only a minority of the station-years is determinate and MOZ is lon/15 h (about 06:05 to 06:35 UTC for the term ‘07 MOZ’), so no whole-hour shift fits and none is made. The time zone audit (GH-2678, 516 station-years at 226 stations) found the same split. Deviates from the source for the real labels: #2711. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: Metadaten_Parameter_tu_termin_01048.txt: “1048;19910101;20010331;…;TT_TER;…(3 Termine: um 07, 14, 21 MOZ, ab 01.01.1987 07:30,14:30,21:30 MEZ)…;Terminwerte in MEZ”; “1048;19600102;19901231;…synoptischen Haupttermine (00,06,12,18 UTC)…;Terminwerte in UTC”; “1048;19340101;19450228;…(3 Termine: 07, 14, 21 MOZ und Tageswerte);Terminwerte in MOZ”; “1048;20010401;20260418;…(3 Termine 06, 12, 18 UTC …);Terminwerte in UTC” – inside https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/subdaily/air_temperature/historical/terminwerte_TU_01048_19340101_20251231_hist.zip (the subdaily directories hold no description PDF).

  • observation/subdaily (cloudiness, moisture, pressure, soil, visibility, wind, wind_extreme): Library: Parsed as UTC (src/wetterdienst/provider/dwd/observation/parser.py); wind_extreme joins the FX3 and FX6 files on the parsed stamp (parser.py). Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: No description PDF in observations_germany/climate/subdaily/air_temperature/ or its historical/ listing; the other subdaily datasets’ zips were not fetched (throttling). Not compared with a second source (#2711): cloudiness, soil, visibility, wind, wind_extreme; the pressure sample of the time zone audit (20 stations, not re-run: UTC 2001+ 15 of 15, MEZ 1991-2000 5 of 6, MEZ 1973-1990 3 of 3 at lag 0) shows the split of the air temperature.

  • observation/daily (climate_summary): Library: MESS_DATUM ‘YYYYMMDD’ padded and parsed as 00:00 UTC of D (src/wetterdienst/provider/dwd/observation/parser.py); no per-parameter window. Data check (windows.py, 2025-04..2026-10): RSK(D) == sum of hourly R1 stamped D 07..D+1 06 in 549/549 days; TXK/TNK fit 00-24 UTC best. Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: Metadaten_Parameter_klima_tag_01048.txt: RSK “06:00 - 06:00 FT. UTC (05:51-05:50 FT.UTC)” (2001-04-01..), “07:30 - 07:30 FT. MEZ (bis 1986 MOZ)” (1991-2001), “06:00 -06:00 FT. UTC” (1960-1990); TXK/TNK “00:00 - 24:00 UTC gemessen” (2001-04-01..), “21:30 VT. - 21:30 MEZ (bis 1986 MOZ)” (1991-2001), “18:00 VT. - 18:00 UTC” (1960-1990); SHK_TAG “06 UTC”; FX “23:51 - 23:50 UTC”; legend “FT = Folgetag” (de: FT = following day, VT = previous day) – inside https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/daily/kl/historical/tageswerte_KL_01048_19340101_20251231_hist.zip. The dataset PDFs https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/daily/kl/DESCRIPTION_obsgermany-climate-daily-kl_en.pdf (and _de) state no day definition. Data check (#2711, 2026-10-16): RSK(D) against the 24 h precipitation of NOAA ISD (AA1 to AA4, period 24 h, minute-00 reports) at 06 UTC, 28 station-years at 15 stations: for every label the date D is the window that closes at 06 UTC of D+1 (ISD of D+1 equals RSK(D), the other days 1 to 2 % exact): ‘06:00 - 06:00 FT. UTC’ 2001+ 88 % exact (2 stations, 248 wet days), ‘07:30 - 07:30 FT. MEZ’ 1987-1990 43 % exact and 73 % within 0.25 mm (3 stations), 1991-2000 39 % and 59 % (12 stations, 2835 wet days), and the same label in periods from 2001 66 % and 81 % (5 stations); the window starts 30 minutes after ISD’s, so it is not exact, and there is no whole day or hour shift. GHCN-daily (key-less, noaa-ghcn-pds) is no second source for most stations: the series of the stations ‘GME…’ (ECA&D, flag E) are DWD’s own values (identical at lag 0 in 95 to 100 % of the days of every label, TXK and TNK one day off in 5 to 9 %); the series of the stations ‘GM…’ (flag S, 3 to 10 stations per label) put TXK and TNK of every label on the same date (28 to 82 % within 0.15 K at lag 0, 4 to 11 % one day off) and RSK of the labels ‘06:00 - 06:00 FT. UTC’ from 2001 and ‘07:30 - 07:30 FT. MEZ’ on the next day (PRCP of D+1 88 % and 55 % within 0.15 mm, of D 36 % and 34 %). The stamp of a daily value is therefore the date D of the window, whichever label it has (‘FT’ windows begin on D, ‘VT’ windows end on D), and no stamp is a day off; the window itself is as the label says to within the 0.5 h that these checks resolve (the time zone audit, not re-run: the start of the TXK/TNK window at -6.0 and -5.5 h for ‘18:00 VT. - 18:00 UTC’, -3.0 and -2.5 h for ‘21:30 VT. - 21:30 MEZ’, 0.0 and +0.5 h for ‘00:00 - 24:00 UTC’). Not compared: the daily datasets other than climate_summary.

  • observation/daily (precipitation_more, solar, temperature_soil, water_equivalent, weather_phenomena, weather_phenomena_more): Library: Parsed as 00:00 UTC of the date (src/wetterdienst/provider/dwd/observation/parser.py). Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: Not fetched (throttling); their Metadaten_Parameter files carry the per-parameter windows as for kl.

  • observation/monthly (climate_summary): Library: MESS_DATUM_BEGINN renamed to timestamp, MESS_DATUM_ENDE dropped (src/wetterdienst/provider/dwd/observation/parser.py), parsed as 00:00 UTC of the 1st (parser.py). Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: “MESS_DATUM_BEGINN begin of interval NUMBER YYYYMMDD”; “MESS_DATUM_ENDE end of interval NUMBER YYYYMMDD” – https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/monthly/kl/DESCRIPTION_obsgermany-climate-monthly-kl_en.pdf; Metadaten_Parameter_klima_monat_01048.txt: “MO_RR;Monatssumme der Niederschlagshoehe;Summe aus Tagessummen” (de: sum of the daily sums). No zone stated.

  • observation/monthly (climate_indices, precipitation_indices, precipitation_more, weather_phenomena): Library: Same code path as climate_summary (src/wetterdienst/provider/dwd/observation/parser.py). Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: Their description PDFs were not fetched (throttling).

  • observation/annual: Library: MESS_DATUM_BEGINN renamed to timestamp, MESS_DATUM_ENDE dropped (src/wetterdienst/provider/dwd/observation/parser.py), parsed as 00:00 UTC (parser.py). Request side: naive start/end are taken as UTC (src/wetterdienst/model/request.py), the window filter is inclusive at both ends (src/wetterdienst/model/values.py); the period choice converts the interval to Europe/Berlin only to pick historical/recent/now files (src/wetterdienst/provider/dwd/observation/api.py). Source: No description PDF found in https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/annual/kl/ or annual/kl/historical/ (listings fetched). The data file header reads “STATIONS_ID;MESS_DATUM_BEGINN;MESS_DATUM_ENDE;QN_4;JA_N;…” with rows “1048;19340101;19341231” (produkt_klima_jahr_19340101_20251231_01048.txt); the monthly PDF defines MESS_DATUM_BEGINN as “begin of interval”. Only kl was looked at.

  • mosmix/hourly: Library: TimeStep strings parsed with to_datetime(‘%Y-%m-%dT%H:%M:%S%.fZ’, time_zone=’UTC’), no shift (provider/dwd/mosmix/api.py; steps read in mosmix/access.py). issue: naive=UTC, aware converted to UTC then floored to the hour (api.py), MOSMIX-L floored to 03/09/15/21 (api.py). Request bounds: naive start/end taken as UTC, aware ones converted to UTC (model/request.py); rows cut with is_between(start, end, closed=’both’) (model/values.py). Source: [de] KML-Formatbeschreibung, https://www.dwd.de/DE/leistungen/met_verfahren_mosmix/kml_formatbeschreibung.pdf?__blob=publicationFile&v=16: “<dwd:IssueTime>2017-04-10T04:00:00.000Z</dwd:IssueTime> <!–Der Herstellungszeitpunkt des Produkts „Lauf“ –>” … “<dwd:TimeStep>2017-04-10T04:00:00.000Z</dwd:TimeStep>” … “TimeStep: Angabe eines Vorhersagezeitpunkts.” (= “statement of a forecast point in time”; the zone is given only by the ISO 8601 ‘Z’ designator, not in words). Real file MOSMIX_L_LATEST_10637.kmz (2026-10-10): “<dwd:IssueTime>2026-10-10T15:00:00.000Z</dwd:IssueTime>”, “<dwd:TimeStep>2026-10-10T16:00:00.000Z</dwd:TimeStep>”. Element table, https://www.dwd.de/DE/leistungen/opendata/help/schluessel_datenformate/kml/mosmix_elemente_xls.xlsx?__blob=publicationFile&v=7: “TTT | Kelvin | Temperature 2m above surface”, “FF | m/s | Wind speed” (no averaging window stated).

  • mosmix/hourly: Library: Same as instant elements: the TimeStep is used unchanged as the UTC timestamp, so a window value carries the stamp of its end (api.py). Verified offline on a real file: TM/RRdc/PEvap/RSunD carry values only at 06 UTC, TX/TN/FXh at 06 and 18 UTC; library returns them at those UTC instants. Source: [de] KML-Formatbeschreibung, https://www.dwd.de/DE/leistungen/met_verfahren_mosmix/kml_formatbeschreibung.pdf?__blob=publicationFile&v=16: “<dwd:IssueTime>2017-04-10T04:00:00.000Z</dwd:IssueTime> <!–Der Herstellungszeitpunkt des Produkts „Lauf“ –>” … “<dwd:TimeStep>2017-04-10T04:00:00.000Z</dwd:TimeStep>” … “TimeStep: Angabe eines Vorhersagezeitpunkts.” (= “statement of a forecast point in time”; the zone is given only by the ISO 8601 ‘Z’ designator, not in words). Real file MOSMIX_L_LATEST_10637.kmz (2026-10-10): “<dwd:IssueTime>2026-10-10T15:00:00.000Z</dwd:IssueTime>”, “<dwd:TimeStep>2026-10-10T16:00:00.000Z</dwd:TimeStep>”. Element table https://www.dwd.de/DE/leistungen/opendata/help/schluessel_datenformate/kml/mosmix_elemente_xls.xlsx?__blob=publicationFile&v=7 (also https://opendata.dwd.de/weather/lib/MetElementDefinition.xml): “RR1c | kg/m2 | Total precipitation during the last hour consistent with significant weather”, “TX | Kelvin | Maximum temperature - within the last 12 hours”, “TM | Kelvin | Mean temperature during the last 24 hours”, “PEvap | kg / m2 | Potential evapotranspiration within the last 24 hours”. Rad1h is only “Global Irradiance” (window not stated).

  • mosmix/hourly (large): Library: Stamped at the 06:00 UTC TimeStep it appears at (api.py), mapped to sunshine_duration_yesterday. Source: https://www.dwd.de/DE/leistungen/opendata/help/schluessel_datenformate/kml/mosmix_elemente_xls.xlsx?__blob=publicationFile&v=7: “SunD | s | Yesterdays total sunshine duration | | x | Sonnenscheindauer Vortag insgesamt”; searched the element table, MetElementDefinition.xml and the KML format PDF: no definition of the day boundary.

  • dmo/hourly: Library: Same KML reader as mosmix; TimeStep parsed with to_datetime(‘%Y-%m-%dT%H:%M:%S.000Z’, time_zone=’UTC’) (provider/dwd/dmo/api.py). Run stamp DDHHMM from the file name dated as the latest real UTC date <= now+1 d (api.py); issue naive=UTC/aware converted, floored to 00/12 UTC (api.py). Request bounds: naive start/end taken as UTC, aware ones converted to UTC (model/request.py); rows cut with is_between(start, end, closed=’both’) (model/values.py). Source: [de] KML-Formatbeschreibung, https://www.dwd.de/DE/leistungen/met_verfahren_mosmix/kml_formatbeschreibung.pdf?__blob=publicationFile&v=16: “<dwd:IssueTime>2017-04-10T04:00:00.000Z</dwd:IssueTime> <!–Der Herstellungszeitpunkt des Produkts „Lauf“ –>” … “<dwd:TimeStep>2017-04-10T04:00:00.000Z</dwd:TimeStep>” … “TimeStep: Angabe eines Vorhersagezeitpunkts.” (= “statement of a forecast point in time”; the zone is given only by the ISO 8601 ‘Z’ designator, not in words). Real file MOSMIX file (2026-10-10): “<dwd:IssueTime>2026-10-10T15:00:00.000Z</dwd:IssueTime>”, “<dwd:TimeStep>2026-10-10T16:00:00.000Z</dwd:TimeStep>”. Real DMO file ptp_gdmog_10637_078_1_101200.kmz: “<dwd:IssueTime>2026-10-10T12:00:00.000Z</dwd:IssueTime>”, “<dwd:TimeStep>2026-10-10T13:00:00.000Z</dwd:TimeStep>” (run stamp 101200 = day 10, 12:00 UTC). No DMO-specific documentation found (the dmo/ directory holds only icon/ and icon-eu/).

  • dmo/hourly: Library: TimeStep used unchanged as UTC timestamp (dmo/api.py). Real 078 run: TX only at 18 UTC, TN only at 06 UTC, returned at those UTC instants. Source: https://www.dwd.de/DE/leistungen/opendata/help/schluessel_datenformate/kml/mosmix_elemente_xls.xlsx?__blob=publicationFile&v=7 (rows outside the MOSMIX columns) and https://opendata.dwd.de/weather/lib/MetElementDefinition.xml: “RR1 | kg/m2 | Total precipitation during the last hour”, “RR3 | kg/m2 | Total precipitation during the last 3 hours”, “RadS3 … Short wave radiation balance during the last 3 hours”, “TX … Maximum temperature - within the last 12 hours”. Zone as in the KML format PDF (‘Z’ designator), https://www.dwd.de/DE/leistungen/met_verfahren_mosmix/kml_formatbeschreibung.pdf?__blob=publicationFile&v=16.

  • road/15_minutes: Library: BUFR year/month/day/hour/minute concatenated and parsed ‘%Y%m%d%H%M’ with time_zone=’UTC’, no shift (provider/dwd/road/api.py); file index stamp ‘-YYMMDDhhmm-’ parsed ‘%y%m%d%H%M’ UTC (api.py) and files cut is_between(start, end) (api.py), then rows again closed=’both’ (model/values.py); naive start/end = UTC (model/request.py). Source: [de] https://www.dwd.de/DE/leistungen/opendata/help/schluessel_datenformate/bufr/bufr_templates_sws_national_pdf.pdf?__blob=publicationFile&v=3: “3 01 011 0 04 001 Jahr … 3 01 012 0 04 004 Stunde … 0 04 005 Minute” and “Dabei ist die Datum/Zeitangabe die “Most typical time for the BUFR message content“”; https://www.dwd.de/DE/leistungen/opendata/help/schluessel_datenformate/bufr/bufr_templates_sws_international_pdf.pdf?__blob=publicationFile&v=5: “3 01 011 Year, month, day / 3 01 012 Hour, minute”. Neither template states UTC (searched both for UTC/time zone). Empirical: KM file swis2-ISXD70_DWKM_102000-2610102000-NW—bin, HTTP Last-Modified ‘Sat, 10 Oct 2026 20:00:18 GMT’, BUFR time 2026-10-10 20:00 in all 253 subsets -> consistent with UTC only (CEST would be 18:00 UTC).

  • road/15_minutes: Library: Message time used unchanged as the UTC timestamp (road/api.py); the 0 04 025 time period is not read, so a longer-than-standard period (flag bit 4 of 0 02 243) would go unnoticed. Source: [de] https://www.dwd.de/DE/leistungen/opendata/help/schluessel_datenformate/bufr/bufr_templates_sws_national_pdf.pdf?__blob=publicationFile&v=3: “0 04 025 Zeitperiode (in der Regel = - 15 Minuten)” before “0 13 011 Niederschlagsmenge / Wasseräquivalent”; “Standardmäßig beträgt die Zeitperiode einer Meldung 15 Minuten.” https://www.dwd.de/DE/leistungen/opendata/help/schluessel_datenformate/bufr/bufr_templates_sws_international_pdf.pdf?__blob=publicationFile&v=5: “0 04 025 Time period or displacement = –15 minutes”. Real KM file: #1#timePeriod = -15 in all 253 subsets.

  • road/15_minutes: Library: Message time used unchanged as UTC timestamp (road/api.py); time periods not read. Source: https://www.dwd.de/DE/leistungen/opendata/help/schluessel_datenformate/bufr/bufr_templates_sws_international_pdf.pdf?__blob=publicationFile&v=5: “0 08 021 Time significance = 2 Time averaged / 0 04 025 Time period or displacement = –10 minutes / 0 11 001 Wind direction / 0 11 002 Wind speed” and “Maximum wind gust / 0 04 025 Time period or displacement In minutes”. Real KM file 2610102000: #2#timePeriod = 0 and #3#timePeriod = missing in all 253 subsets.

  • swsmos/hourly: Library: YYYYMMDDHHmm parsed ‘%Y%m%d%H%M’ and labelled UTC with replace_time_zone(‘UTC’), no shift (provider/dwd/swsmos/api.py). issue naive=UTC/aware converted, floored to the hour (api.py); run file names swsmos_<YYYYMMDDHH0000> read as UTC (api.py). Request bounds: naive start/end taken as UTC, aware ones converted to UTC (model/request.py); rows cut with is_between(start, end, closed=’both’) (model/values.py). Source: [de] https://www.dwd.de/DE/leistungen/swis_swsmos/swis_swsmos_beschreibung.pdf?__blob=publicationFile&v=2: “2) YYYYMMDDHHm: Datum + Vorhersagezeitpunkt [UTC]” (= date + forecast time [UTC]); “3) TL: Lufttemperatur [°C]”, “13) TS: Temperatur der Fahrbahnoberfläche [°C]”. Real run swsmos_20261010190000: line 2 ‘202610101900’, first row ‘A006;…;202610102000;9.2;…’, Last-Modified 19:57:01 GMT.

  • swsmos/hourly: Library: Same stamp used unchanged (swsmos/api.py), so window values carry their end. Source: [de] https://www.dwd.de/DE/leistungen/swis_swsmos/swis_swsmos_beschreibung.pdf?__blob=publicationFile&v=2: “2) YYYYMMDDHHm: Datum + Vorhersagezeitpunkt [UTC]”; “5) RRL1c: Niederschlag flüssig der letzten Stunde [mm]” (= liquid precipitation of the last hour); “7) RR6: Niederschlag flüssig der letzten 6 Stunden akkumuliert [mm]”; “11) R650: Niederschlag flüssig+fest mehr als 5mm der letzten 6 Stunden Wahrscheinlichkeit [%]”.

  • poi/hourly: Library: Columns 1+2 ‘DD.MM.YY HH:MM’ parsed and labelled UTC with replace_time_zone(‘UTC’), no shift (provider/dwd/poi/api.py). Request bounds: naive start/end taken as UTC, aware ones converted to UTC (model/request.py); rows cut with is_between(start, end, closed=’both’) (model/values.py). Verified offline on the real file: 25 hourly rows 09.10.26 19:00 .. 10.10.26 19:00 -> 2026-10-09 19:00 .. 2026-10-10 19:00 UTC, Datetime(us, UTC). Source: File header itself (https://opendata.dwd.de/weather/weather_reports/poi/10637-BEOB.csv (fetched 2026-10-10 ~19:40 UTC); no POI format document found on the DWD help page besides the present-weather code table): line 1 “surface observations;Parameter description;…”, line 3 [de] “Datum;Uhrzeit (UTC);…;Temperatur (2m);…” (= date; time (UTC)). Wind: the English header names the columns “mean_wind_speed_during last_10_min_at_10_meters_above_ground” and “mean_wind_direction_during_last_10 min_at_10_meters_above_ground” (the German header says only “Windgeschwindigkeit” and “Windrichtung”), the mean over the last 10 minutes of the report, so the stamp ends their window and they are not instants. Instant nature of the other columns is not stated beyond their names.

  • poi/hourly: Library: Report time used unchanged as UTC timestamp (poi/api.py). Real file: TX12h at 06 and 18 UTC, RR24h at 06 UTC, and the column “Maximalwind (vergangener Tag)” (maximum_of_10_minutes_mean_of_wind_speed_for_previous_day) at 00, 06, 12 and 18 UTC with a value equal to the maximum of the 6 hourly values of “Maximalwind (letzte Stunde)” up to the stamp, the report’s own hour included (read 2026-10-15 20:29 UTC: 13 km/h at 18:00 against 13, 13, 11, 8, 7, 3 at 13:00 to 18:00; 12 km/h at 12:00 against 4, 5, 6, 9, 9, 12 at 07:00 to 12:00, where 06:00 to 11:00 would give 9; 7 and 6 at 06 and 00 likewise), so it is named wind_speed_rolling_mean_max_last_6h (GH-2817). Source: https://opendata.dwd.de/weather/weather_reports/poi/10637-BEOB.csv (fetched 2026-10-10 ~19:40 UTC), line 3 [de]: “Uhrzeit (UTC)”, “Niederschlag (letzte Stunde)”, “Niederschlag (letzte 24 Stunden)”, “Maximumtemperatur (letzte 12 Stunden, 2m)”, “Windboen (letzte 6 Stunden)”, “Globalstrahlung (letzte Stunde)” (letzte = last).

  • poi/hourly: Library: Stamped at the 06:00 UTC report row (poi/api.py) under *_yesterday names; real file: daily_mean_of_temperature_previous_day only at 10.10.26 06:00 -> 2026-10-10 06:00 UTC. Source: https://opendata.dwd.de/weather/weather_reports/poi/10637-BEOB.csv (fetched 2026-10-10 ~19:40 UTC), line 3 [de]: “mittlere Temperatur (vergangener Tag, 2m)”, “Maximumtemperatur (vergangener Tag, 2m)”, “Sonnenscheindauer (vergangener Tag)”, “Windboen (vergangener Tag)” (vergangener Tag = previous day). Searched the DWD opendata help page for a POI description: none besides poi_present_weather_zuordnung_pdf.pdf.

  • phenology/annual: Library: timestamp = pl.datetime(Referenzjahr, 1, 1, time_zone=’UTC’), value = Jultag (provider/dwd/phenology/api.py); period choice derives year boundaries in Europe/Berlin (api.py); station end date ‘%d.%m.%Y’ as UTC midnight (api.py). Request bounds: naive start/end taken as UTC, aware ones converted to UTC (model/request.py); rows cut with is_between(start, end, closed=’both’) (model/values.py). Source: https://opendata.dwd.de/climate_environment/CDC/observations_germany/phenology/annual_reporters/fruit/recent/DESCRIPTION_obsgermany-phenology-annual_reporters-fruit-recent_en.pdf: “Referenzjahr | year corresponding to phase”, “Eintrittsdatum | date of observation | yyyymmdd”, “Jultag | date of observation | day of the year”. No time zone stated (dates only). Real PH_Jahresmelder_Landwirtschaft_Kulturpflanze_Winterweizen_akt.txt: in 2023/2025/2026 Jultag == day-of-year of Eintrittsdatum; in 2024 Jultag == day-of-year - 1 for 3359 rows dated March-December (0 for 5 Jan/Feb rows); 2 rows (Referenzjahr 2024) have Eintrittsdatum in 2023.

  • radar/n/a: Library: File-name stamps (YYMMDDhhmm / YYYYMMDDhhmm[ss] / YYYYMM) parsed and labelled UTC (provider/dwd/radar/index.py; api.py), the run of a composite written YYYYMMDD_HHMM and that of KONRAD3D YYYYMMDDThhmm[ss], its lead time not added. start/end: naive = UTC, aware converted to UTC (api.py), snapped to the product cadence, RADOLAN_CDC to hh:50 (api.py), file index cut is_between(start, end, closed=’both’) (api.py). The lead time of a forecast file (RV, WN, RS) is not added: every lead-time file of one run carries the base time. Source: [de] https://www.dwd.de/DE/leistungen/radolan/radolan_info/radolan_radvor_op_komposit_format_pdf.pdf?__blob=publicationFile&v=26: “3I2 Zeitpunkt der Messung: Tag, Stunde und Minute (ddhhmm) in UTC” (= time of measurement … in UTC); “Bei diesen Vorhersageprodukten wird mit der Datum-Zeit-Gruppe YYMMDDhhmm die Zeit angegeben, von dem Produkt die Vorhersage aus berechnet wird. Mit der Headerkennung VV bzw. des Dateinamensbestandteils vvv wird das Ende der vorhergesagten Zeit in Minuten nach der in der Datum-Zeit-Gruppe angegebenen Zeit definiert.” (= for these forecast products the date-time group gives the time from which the forecast is computed; VV/vvv defines the end of the forecast time in minutes after it); RH: “auf eine Stunde aufsummiert” (summed over one hour), SF: “24h-aufsummiertes Endergebnis” – start/end of the window vs the stamp not stated. KONRAD3D (Kurzdokumentation, https://www.dwd.de/DE/leistungen/radarprodukte/konrad3d_kurzbeschreibung.pdf?__blob=publicationFile): “Jedem Lauf ist eine Referenzzeit t zugeordnet, die auch im Zeitstempel im Dateinamen enthalten ist” (= each run has a reference time t, which is also the time stamp in the file name) and “Der Referenzzeitstempel von KONRAD3D bezieht sich also per Definition auf den Beginn des Radarscanzyklus” (= the reference time refers by definition to the start of the radar scan cycle).

  • alerts/n/a: Library: CAP stamps parsed with datetime.fromisoformat and converted with astimezone(UTC) (provider/dwd/alerts/parser.py), so the offset decides; columns Datetime(us, UTC) (alerts/api.py). Snapshot names Z_CAP_C_EDZW_<yyyyMMddHHmmss> read as UTC (api.py) and the newest one <= the requested timestamp chosen (api.py); a naive request timestamp is taken as UTC (api.py). Source: https://www.dwd.de/DE/leistungen/opendata/help/warnungen/cap_dwd_profile_en_pdf_2_1_14.pdf?__blob=publicationFile&v=3: “2.2.2.10 <onset> (ASSURED) Info The expected time of the beginning of the event Format <A> + <B> <A> Date and time (in CET/CEST) <yyyy>-<MM>-<dd>T<HH>:<mm>:<ss> <B> Time offset to UTC (UTC: +00:00; CET: +01:00; CEST: +02:00) Example 2010-05-18T12:24:18+02:00 (= 2010-05-18T10:24:18+00:00)”; “2.2.2.11 <expires> … The expected end of the event”; “2.2.2.9 <effective> … Time at which the warning was issued”. Snapshot file name zone not documented (https://www.dwd.de/DE/leistungen/opendata/help/warnungen/dwd_warnings_products_overview_en_pdf.pdf?__blob=publicationFile&v=4: “Z_CAP_C_EDZW_<yyyyMMddHHmmss>_PVW_STATUS_”); empirically UTC: Z_CAP_C_EDZW_20261010200500_…_DISTRICT_DE.zip listed with mtime 10-Oct-2026 20:05:01 (server listing in GMT).

  • derived/hourly (radiation_global): Library: MESS_DATUM ‘YYYYMMDDHH’ + ‘00’ parsed as UTC and rounded to the hour (a no-op) (src/wetterdienst/provider/dwd/derived/parser.py); FG and FG_UN files joined on it (parser.py). Request side: naive start/end taken as UTC (src/wetterdienst/model/request.py); window filter inclusive at both ends (src/wetterdienst/model/values.py). Source: “The specified MESS_DATUM describes the end of the synoptic hour for which the data apply. The synoptic hour HH:00 denotes the hour from (HH-2):50 to (HH-1):50.” – https://opendata.dwd.de/climate_environment/CDC/derived_germany/climate/hourly/duett/DESCRIPTION_derived_germany-climate-hourly-duett_en.pdf; Metadaten_Parameter_fg_duett_18088.txt: “FG_DUETT;stdl. Globalstrahlung (berechnet DUETT-Raster);…;in UTC”. Data: FG_DUETT(HH) vs station GS_10 summed over the hour ending HH-10min corr 0.9924/MAE 4.6 J/cm2, ending HH 0.9899/5.9.

  • derived/hourly (sunshine_duration): Library: As radiation_global (src/wetterdienst/provider/dwd/derived/parser.py). Request side: naive start/end taken as UTC (src/wetterdienst/model/request.py); window filter inclusive at both ends (src/wetterdienst/model/values.py). Source: Same DUETT PDF: “The synoptic hour HH:00 denotes the hour from (HH-2):50 to (HH-1):50.” – https://opendata.dwd.de/climate_environment/CDC/derived_germany/climate/hourly/duett/DESCRIPTION_derived_germany-climate-hourly-duett_en.pdf

  • derived/daily (soil): Library: ‘datum’ parsed ‘%Y%m%d’ as 00:00 UTC (src/wetterdienst/provider/dwd/derived/parser.py); periods/unique (src/wetterdienst/provider/dwd/derived/api.py). Request side: naive start/end taken as UTC (src/wetterdienst/model/request.py); window filter inclusive at both ends (src/wetterdienst/model/values.py). Source: Silent: https://opendata.dwd.de/climate_environment/CDC/derived_germany/soil/daily/historical/BESCHREIBUNG_derived_ermany_soil_daily_historical_de_v2.pdf (“Statistische Verarbeitung Tageswert, Zeitreihe, Tagessumme, Tagesmittel”; no zone or day definition).

  • derived/monthly (soil): Library: ‘monat’ parsed ‘%Y%m’ as 00:00 UTC of the 1st (src/wetterdienst/provider/dwd/derived/parser.py). Request side: naive start/end taken as UTC (src/wetterdienst/model/request.py); window filter inclusive at both ends (src/wetterdienst/model/values.py). Source: Silent: https://opendata.dwd.de/climate_environment/CDC/derived_germany/soil/monthly/historical/DESCRIPTION_derived_germany-soil-monthly-historical-v2_en.pdf

  • derived/monthly (heating_degreedays, cooling_degreehours_13, cooling_degreehours_16, cooling_degreehours_18): Library: Month list = the UTC calendar months from the one the request’s start falls in to the one its end falls in, each as its first day at 00:00 UTC, stepped ‘1mo’ (src/wetterdienst/provider/dwd/derived/api.py, _first_day_of_month_in_utc); each value is stamped with that datetime (api.py) and the file month is its strftime(‘%Y%m’) (api.py). The window cut (src/wetterdienst/model/values.py) keeps a month only if its stamp lies within start and end, so a start of 2025-01-01T06:30 does not return January, and a start of 2025-01-01T00:00+01:00 (2024-12-31 23:00 UTC) fetches December but returns January on. Until GH-2679 the stamp kept the start’s time of day and offset (a start of 06:30 gave 2025-01-01 06:30 UTC), and a start of 2025-01-01T00:00+01:00 fetched December to February for a January to March request and stamped them 2024-12-31 23:00, 2025-01-31 23:00 and 2025-02-28 23:00 UTC. Source: “Monat Year and month yyyymm” – https://opendata.dwd.de/climate_environment/CDC/derived_germany/techn/monthly/heating_degreedays/hdd_3807/recent/DESCRIPTION_derivgermany_techn_monthly_heating_degreedays_hdd_3807_recent_en.pdf and https://opendata.dwd.de/climate_environment/CDC/derived_germany/techn/monthly/cooling_degreehours/DESCRIPTION_derivgermany_techn_monthly_cooling_degreehours_cdh_recent_en.pdf; file header “#Auswertezeitraum 01-JAN-2025 bis 31-JAN-2025” (gradtage_202501.csv). No zone stated.

  • derived/monthly (climate_correction_factor): Library: As heating_degreedays: stamped with the first day of the month at 00:00 UTC that names its file (src/wetterdienst/provider/dwd/derived/api.py); file KF_<start>_<start+1y-1d>.csv (api.py). Until GH-2679 the stamp and the file named by it followed the start’s time of day and offset, as for heating_degreedays. Source: Directory listing https://opendata.dwd.de/climate_environment/CDC/derived_germany/techn/monthly/climate_correction_factor/recent/ (e.g. KF_20250801_20260731.csv); DESCRIPTION_derivgermany_techn_monthly_climate_correction_factor_recent_en.pdf states neither zone nor window semantics beyond ‘calculation period’.

eccc#

network

resolution

datasets

source time zone

reference time

verified

observation

hourly

all

local standard time (fixed offset, no DST)

mixed (instant hourly observations; precip_amount covers ‘minutes 00 through 60’ of an hour whose relation to the stamp the docs do not state)

partly

observation

daily

all

UTC-bounded climatological day (06:01 UTC to 06:00 UTC) at principal stations; calendar day at once-daily stations

start (the date names the climatological day beginning 0601 UTC that date)

partly

observation

monthly

all

unverified (months of daily values; the daily day definition applies)

start (month label)

partly

Notes:

  • observation/hourly: Library: stamped from the `UTC_DATE` field of the row (`LOCAL_DATE` plus the station’s standard offset all year, so Toronto is +5 h in July as well), read as UTC (src/wetterdienst/provider/eccc/observation/api.py); `LOCAL_DATE` is not read for this resolution. The station list’s `HLY_FIRST_DATE`/`HLY_LAST_DATE` (the station’s `start_timestamp`/`end_timestamp` for this resolution, GH-2747; a station without a `HLY_FIRST_DATE` is not listed for hourly, GH-2867) are read as LST too: `HLY_FIRST_DATE` of Toronto Intl A, ‘2013-06-11 09:00:00’, is the `LOCAL_DATE` of its first hourly row, `UTC_DATE` 14:00; the catalogue’s `HLY_FIRST_DATE`/`HLY_LAST_DATE` of four more stations and `LAST_DATE` of one closed station (Tasiujaq) match the first and last hourly `LOCAL_DATE` the same way, while `HLY_LAST_DATE` (and `LAST_DATE`) of an active station is a catalogue refresh stamp (‘2026-10-10 12:30:12’ for 918 of them) whose clock is not established. The dates are converted with the fixed offset of the `TIMEZONE` code (EST -5 h, CST -6, MST -7, PST -8, AST -4, NST -3:30) (GH-2693). It was once localised with the station’s *civil* zone, which stamped every value 1 h early during DST (GH-2675). A year is requested as `UTC_YEAR=<year>`, the year of the stamps (api.py), the request window is applied afterwards, both ends inclusive. A request without start and end asks for the years from that of the first hourly row (`sortby=UTC_DATE&limit=1`) to that of the last (`sortby=-UTC_DATE`), as the station list’s dates follow the temperature and not every series (GH-2820, GH-2864). The `datetime` filter of the API applies to `LOCAL_DATE` and is not used for this resolution. Source: “The Local Standard Time(LST) is used for observation purposes and is that of the standard time zone in which the station is located, whether or not “daylight saving time” is adopted for other purposes. In Canada, Local Standard Time is commonly used for archiving surface weather observations.” and “Hourly: indicates data values for observations taken on an hourly basis.” and “The total hourly precipitation (element 262) is the total precipitation amount for minutes 00 through 60, inclusive, computed as the sum of the four 15-minute precipitation amounts.” (https://climate.weather.gc.ca/glossary_e.html); bulk CSV header “Date/Time (LST)” (https://climate.weather.gc.ca/climate_data/bulk_data_e.html?format=csv&stationID=51459&Year=2025&Month=7&Day=1&timeframe=1); the API returns LOCAL_DATE ‘2025-07-01 12:00:00’ with UTC_DATE ‘2025-07-01T17:00:00’ for TORONTO INTL A in July (https://api.weather.gc.ca/collections/climate-hourly/items?STN_ID=51459&datetime=2025-07-01T10:00:00/2025-07-01T12:00:00&f=json).

  • observation/daily: Library: `LOCAL_DATE` (‘YYYY-MM-DD 00:00:00’) is a calendar date and is stamped at 00:00 UTC of it, whatever the station’s zone (src/wetterdienst/provider/eccc/observation/api.py); it was localised as the station’s civil midnight, Toronto 05:00 UTC in winter and 04:00 UTC in summer, until GH-2683. A year is requested as the calendar year of `LOCAL_DATE` (api.py). Request window applied afterwards, both ends inclusive, so start=end=D returns D. Source: “The climatological day tends to start and end at the same hour of two consecutive days. For example, the climatological day begins at the 0601 UTC observations on Day 1 and ends at the 0600 UTC observation on Day 2.” and “For sites reporting only once per day, the calendar day rather than climatological day applies.” and “At most principal stations the maximum and minimum temperatures are for a day beginning at 0601 Greenwich (or Universal) Mean Time, which is within a few hours of midnight local standard time in Canada.” (https://climate.weather.gc.ca/glossary_e.html)

  • observation/monthly: Library: `LOCAL_DATE` (‘YYYY-MM’) is parsed as the 1st 00:00 and stamped at 00:00 UTC of it, whatever the station’s zone (src/wetterdienst/provider/eccc/observation/api.py); it was localised as the station’s civil midnight, Toronto 05:00 UTC Nov-Mar and 04:00 UTC Apr-Oct, until GH-2683. Source: “Monthly: are averages for each month, derived from daily data values.” (https://climate.weather.gc.ca/glossary_e.html). No statement of a zone for the month boundaries found in the glossary.

fmi#

network

resolution

datasets

source time zone

reference time

verified

observation

hourly

data

UTC

mixed (t2m/td/rh/p_sea/vis: value at the stamp, t2m ‘avg PT1M’; ws/wd/wg ‘PT10M’ and r_1h ‘acc PT1H’: stamp position unstated (observed: end, the 10 minutes and the hour before the stamp))

partly

observation

daily

data

UTC

unverified (observed: tmax/tmin the 24 h 18 UTC D-1 to 18 UTC D, rrday 06 UTC D to 06 UTC D+1, labelled D)

partly

Notes:

  • observation/hourly (data): Library: parser.py parses ‘Time’ (…Z) as UTC; request bounds astimezone(UTC) and sent as …Z (api.py) in 168 h windows sharing boundaries, de-duplicated on (parameter, timestamp) (api.py); framework trims inclusively. Source: “What time zone is used in the open data? UTC time is used in the open data and in the Download observations service as well.” (https://en.ilmatieteenlaitos.fi/open-data-questions); ObservableProperty metadata: t2m ‘avg’ ‘PT1M’, r_1h ‘acc’ ‘PT1H’, ws_10min ‘avg’ ‘PT10M’, wg_10min ‘max’ ‘PT10M’ (https://opendata.fmi.fi/meta?observableProperty=observation&param=t2m,r_1h,ws_10min,wg_10min&language=eng). Searched both FAQ pages, the stored-query descriptions and the WFS manual: no statement whether aggregates are stamped at start or end. Data check (not documentation), Helsinki-Vantaa airport 100968, 2025-06-01 to 08, the 10-minute series of the same WFS (timestep=10): r_1h at H equals the sum of ri_10min/6 stamped H-50..H in 9 of 9 wet hours (H..H+50: 1 of 9), so the hour is the one before the stamp; t2m of the hourly series against NOAA GHCNh Helsinki-Seutula, 457 hours of 2025-06-01 to 20: mean absolute difference 0.04 K at the stamp, 0.85 K one hour either side.

  • observation/daily (data): Library: parser.py parses ‘Time’ (e.g. 2025-03-30T00:00:00Z) as UTC unchanged; bounds astimezone(UTC) (api.py); 365-day windows; framework trims inclusively, so a mid-day start excludes that day’s 00:00 row. Source: “What time zone is used in the open data? UTC time is used in the open data and in the Download observations service as well.” (https://en.ilmatieteenlaitos.fi/open-data-questions); stored query ‘fmi::observations::weather::daily::simple’ says only ‘Daily weather observations from weather stations.’ and the metadata gives tday ‘avg P1D’, tmax/tmin ‘max’/’min’ ‘PT24H’, rrday ‘sum’ ‘PT24H’, snow ‘instant’ ‘P1D’ — no window. Empirical (Helsinki 100971, 2025-03-29..31,): rrday labelled D equals the 06 UTC D -> 06 UTC D+1 sum of r_1h (0.3/0.9~0.8/0.0); tmin labelled 2025-03-31 (4.9) is above the UTC-day 10-min minimum (4.4), so tmin is not a 00-24 UTC value; tday is closest to the UTC-day mean. Not a documented definition.

imgw#

network

resolution

datasets

source time zone

reference time

verified

hydrology

daily

hydrology

UTC

unverified (observed: mixed (stage/discharge at automatic stations: mean over the UTC day D; at observer stations and temperature_water: instant 06 UTC D))

partly

hydrology

monthly

hydrology

unverified (daily values UTC-based, see daily row; month bounds not stated)

start (calendar month; mean = mean of daily values, min/max = extremes of all observations in the month)

partly

meteorology

daily

synop

UTC

start (UTC day D 00-24 UTC: daily means = mean of the 8 terms 00..21 UTC; stamp D 00:00 UTC)

yes

meteorology

daily

synop

UTC

start (precipitation_amount: 06:01 UTC D to 06:00 UTC D+1, dated D; precipitation_amount_day 06-18 UTC D, _night 18 UTC D-06 UTC D+1, dated D)

yes

meteorology

daily

synop

UTC

end (temperature_air_max_2m/min_2m/min_0_05m: 18:01 UTC D-1 to 18:00 UTC D, dated D)

yes

meteorology

daily

synop

UTC

instant (snow_depth: 06 UTC of D)

yes

meteorology

daily

climate

unverified (terms GG 06/12/18 in k_t; their zone is not stated in k_t_format.txt or Opis.txt)

mixed (precipitation_amount 06 UTC D to 06 UTC D+1 dated D; temperature_air_mean_2m = (T06+T18+Tmax+Tmin)/4 of D; Tmax/Tmin window unverified)

partly

meteorology

daily

precipitation

UTC (06 UTC reading, per IMGW bulletin)

start (precipitation_amount 06 UTC D to 06 UTC D+1, dated D; snow_depth/snow_depth_new reading time unverified)

partly

meteorology

monthly

synop

UTC

start (calendar month = aggregate of the daily rows dated in the month; precipitation thus 06 UTC on the 1st to 06 UTC on the 1st of the next month)

partly

meteorology

monthly

climate, precipitation

unverified

unverified (observed: start (calendar month; month bounds of the underlying days as in the daily rows – unverified for these datasets))

no

Notes:

  • hydrology/daily (hydrology): Library: hydrology/api.py maps the hydrological year/month index to calendar year/month (Nov=01 -> year-1, month 11) and builds pl.datetime(year, month, day) 00:00 replace_time_zone(‘UTC’); file pruning widened by 2 months (api.py) and per-file ranges from the hydrological year (api.py). Mapping checked against the file’s own calendar-month field COMSCK: 351,742/351,742 rows agree (codz_2022_01, codz_2024). Source: Rocznik Hydrologiczny 2024 (pl), https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/Roczniki/Rocznik%20hydrologiczny/Rocznik%20Hydrologiczny%202024.pdf: ‘Dla stacji, w których prowadzona jest automatyczna rejestracja stanów wody wartościami dobowymi są wartości średnie chronologiczne z pomiarów rejestrowanych co 10 minut. 2. Dla stacji, w których pomiary wykonuje wyłącznie obserwator część tabeli z wartościami dobowymi zawiera stany wody z godziny 6 UTC.’ / ‘wartości dobowe przepływu odpowiadają stanom wody z godziny 6 UTC’ / ‘dobowych wartości temperatury, które pochodzą z pomiarów wykonywanych o godzinie 6 UTC’ [Translation: automatic stations: daily = chronological mean of 10-min records; observer stations: the 06 UTC reading (discharge from the 06 UTC stage); water temperature: measured at 06 UTC.] The day bounds of the mean are NOT stated. Checked: codz_2024 March 2024 stage equals the mean of IMGW 10-min telemetry (Arch/Telemetria/Hydro/2024/Hydro_2024-03.zip) over the telemetry’s calendar day in 94.6% of 20,360 station-days, 06-06 UTC 48.6%, Warsaw civil day 82.7%; telemetry clock is UTC: Meteo telemetry B00300S at Warszawa-Okecie matches the UTC s_t temperatures at offset 0 (mean |d| 0.10 K vs 0.68-0.71 at +-1 h), and hydro telemetry has 02:00-02:50 on 2024-03-31 (no Warsaw civil time). Hydrological year: ‘Rok hydrologiczny zaczyna się 1 listopada poprzedniego roku kalendarzowego, a kończy 31 października’ (https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/dane_hydrologiczne/hydrologia_info_ogolne.txt).

  • hydrology/monthly (hydrology): Library: hydrology/api.py hydrological->calendar month shift pl.datetime(year, month, 1) replace_time_zone(‘UTC’); extremity 1/2/3 -> min/mean/max (api.py); request bounds as above (inclusive). Source: MIES_publiczne_format.txt (https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/dane_hydrologiczne/miesieczne/MIES_publiczne_format.txt): ‘MCWSKEX - Wskaźnik ekstremum (1 - minimum, 2 - średnia, 3 - maksimum)’ … ‘MCMSCK - Miesiąc kalendarzowy’ [extremity indicator 1 min/2 mean/3 max; calendar month]; Rocznik Hydrologiczny 2024 (https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/Roczniki/Rocznik%20hydrologiczny/Rocznik%20Hydrologiczny%202024.pdf): ‘W wyborze stanów ekstremalnych uwzględnione są wszystkie obserwacje wykonane na stacji.’ [all observations at the station are used to select extremes]. No time zone stated. Checked: CHAŁUPKI 149180020 Nov 2023 mean 141 = mean of daily stages 140.67; min 104 / max 179 lie outside the daily range 109..171 (sub-daily extremes).

  • meteorology/daily (synop): Library: src/wetterdienst/provider/imgw/meteorology/api.py builds the stamp with pl.datetime(year, month, day) (00:00, naive) replace_time_zone(‘UTC’) – no zone conversion, no shift. Request bounds: naive start/end taken as UTC (model/request.py), rows kept with is_between(start, end, closed=’both’) (model/values.py); file pruning by UTC month/year ranges (meteorology/api.py, fileindex.py). Source: Rocznik Meteorologiczny 2024 (pl), https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/Roczniki/Rocznik%20meteorologiczny/Rocznik%20Meteorologiczny%202024.pdf, sec. 1.1/2.2: ‘Do opracowania Rocznika Meteorologicznego wykorzystano dane z 8 terminów pomiarowo-obserwacyjnych, przypadających na godziny: 00, 03, 06, 09, 12, 15, 18 i 21 czasu uniwersalnego UTC.’ / ‘Wartości średnie dobowe (śr) elementów meteorologicznych wyznaczono jako średnie arytmetyczne z 8 wartości terminowych o godzinie: 00, 03, 06, 09, 12, 15, 18 i 21 UTC’ [Translation: the 8 observation terms are at 00..21 UTC; daily means are the arithmetic mean of these 8 terms.] Checked on the archive files: Warszawa-Okecie 2024 s_d STD 366/366 and s_d_t TEMP 182/182 days equal the mean of the 00..21 UTC terms of the same date in s_t. Applies to temperature_air_mean_2m, humidity_relative, wind_speed, cloud_cover_total, pressure_*.

  • meteorology/daily (synop): Library: src/wetterdienst/provider/imgw/meteorology/api.py builds the stamp with pl.datetime(year, month, day) (00:00, naive) replace_time_zone(‘UTC’) – no zone conversion, no shift. Request bounds: naive start/end taken as UTC (model/request.py), rows kept with is_between(start, end, closed=’both’) (model/values.py); file pruning by UTC month/year ranges (meteorology/api.py, fileindex.py). The stamp D 00:00 UTC lies 6 h before the period begins (same calendar date; no day shift). Source: Rocznik Meteorologiczny 2024 (pl), https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/Roczniki/Rocznik%20meteorologiczny/Rocznik%20Meteorologiczny%202024.pdf: ‘Sumy dobowe opadu odnoszą się do umownej doby opadowej, trwającej od godziny 06.01 UTC dnia bieżącego do godziny 06.00 UTC dnia następnego.’ [Translation: daily precipitation sums refer to the conventional precipitation day from 06:01 UTC of the current day to 06:00 UTC of the next day.] Checked: s_d SMDB equals the sum of the 6-h amounts at 12,18 UTC D and 00,06 UTC D+1 on 145/145 wet days (end-day dating: 5/145); s_d_t WODZ = 12+18 UTC D 63/63, WONO = 00+06 UTC D+1 59/60.

  • meteorology/daily (synop): Library: src/wetterdienst/provider/imgw/meteorology/api.py builds the stamp with pl.datetime(year, month, day) (00:00, naive) replace_time_zone(‘UTC’) – no zone conversion, no shift. Request bounds: naive start/end taken as UTC (model/request.py), rows kept with is_between(start, end, closed=’both’) (model/values.py); file pruning by UTC month/year ranges (meteorology/api.py, fileindex.py). The stamp D 00:00 UTC lies inside the period (which began 18:01 UTC D-1). Source: Rocznik Meteorologiczny 2024 (pl), https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/Roczniki/Rocznik%20meteorologiczny/Rocznik%20Meteorologiczny%202024.pdf: ‘Maksima i minima dobowe odnoszą się do umownej doby termicznej, trwającej od godziny 18.01 UTC dnia poprzedniego do godziny 18.00 UTC dnia bieżącego.’ and for Tg ‘Odnosi się ona do zdefiniowanej wyżej doby termicznej.’ [Translation: daily maxima/minima refer to the thermal day from 18:01 UTC of the previous day to 18:00 UTC of the current day; the ground minimum refers to the same day.] Checked: s_d TMAX/TMIN = max/min of the 12-h extremes reported at 06 and 18 UTC D on 366/366 days.

  • meteorology/daily (synop): Library: src/wetterdienst/provider/imgw/meteorology/api.py builds the stamp with pl.datetime(year, month, day) (00:00, naive) replace_time_zone(‘UTC’) – no zone conversion, no shift. Request bounds: naive start/end taken as UTC (model/request.py), rows kept with is_between(start, end, closed=’both’) (model/values.py); file pruning by UTC month/year ranges (meteorology/api.py, fileindex.py). Stamp D 00:00 UTC is 6 h before the reading. Source: Rocznik Meteorologiczny 2024 (pl), https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/Roczniki/Rocznik%20meteorologiczny/Rocznik%20Meteorologiczny%202024.pdf: ‘Dobowa jej grubość (Sh) jest warstwą skumulowanej pokrywy śnieżnej, mierzonej codziennie o godzinie 06 UTC.’ [Translation: the daily snow depth is measured every day at 06 UTC.] Checked: s_d PKSN equals the 06 UTC s_t PKSN of the same date on 22/23 snow days.

  • meteorology/daily (climate): Library: src/wetterdienst/provider/imgw/meteorology/api.py builds the stamp with pl.datetime(year, month, day) (00:00, naive) replace_time_zone(‘UTC’) – no zone conversion, no shift. Request bounds: naive start/end taken as UTC (model/request.py), rows kept with is_between(start, end, closed=’both’) (model/values.py); file pruning by UTC month/year ranges (meteorology/api.py, fileindex.py). Source: IMGW Informator Meteorologiczny 2025-06A (pl), https://cmm.imgw.pl/wp-content/uploads/production/informator/202506A.pdf: ‘*Pomiar opadu wykonywany jest o godz. 6:00 UTC (dla Polski lokalny czas zimowy +1 godz., lokalny czas letni +2 godz.) i obejmuje 24 godz. okres – od godz. 6:00 UTC dnia poprzedzającego pomiar do godz. 6:00 UTC w dniu wykonania pomiaru. Po wykonaniu pomiaru opadu jego wysokość zapisana zostaje pod datą dnia poprzedzającego’ [Translation: precipitation is measured at 06:00 UTC and covers 06:00 UTC of the previous day to 06:00 UTC of the measurement day; it is recorded under the previous day’s date.] NB the same bulletin elsewhere says a daily sum ‘zapisywana jest pod datą końcową okresu sumowania’ (recorded under the END date) – contradictory. Format files k_d_format.txt / k_d_t_format.txt / Opis.txt (https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/dane_meteorologiczne/) are silent on time zone and day bounds. Checked: k_d_t TEMP and k_d STD = (T06+T18+TMAX+TMIN)/4 from k_t on 1488/1488 station-days (March 2024); 2024 daily precipitation of LEGIONOWO (27 km) correlates with the Warszawa-Okecie synop at lag 0 r=0.90 vs 0.05/0.01 at lag -1/+1, 5 of 6 climate stations at 51-52N 20-21E peak at lag 0.

  • meteorology/daily (precipitation): Library: src/wetterdienst/provider/imgw/meteorology/api.py builds the stamp with pl.datetime(year, month, day) (00:00, naive) replace_time_zone(‘UTC’) – no zone conversion, no shift. Request bounds: naive start/end taken as UTC (model/request.py), rows kept with is_between(start, end, closed=’both’) (model/values.py); file pruning by UTC month/year ranges (meteorology/api.py, fileindex.py). Absent days are filled as 0 with quality_flags 11 on naive dates (meteorology/api.py) – no zone involved. Source: IMGW Informator Meteorologiczny 2025-06A (pl), https://cmm.imgw.pl/wp-content/uploads/production/informator/202506A.pdf: ‘*Pomiar opadu wykonywany jest o godz. 6:00 UTC … i obejmuje 24 godz. okres – od godz. 6:00 UTC dnia poprzedzającego pomiar do godz. 6:00 UTC w dniu wykonania pomiaru. Po wykonaniu pomiaru opadu jego wysokość zapisana zostaje pod datą dnia poprzedzającego’ [Translation: measured at 06 UTC over the preceding 24 h and recorded under the previous day’s date] (same bulletin contradicts this on another page, see climate row). o_d_format.txt is silent on time. Checked: all 22 o_d stations at 51-52N 20-21E with >=300 days in 2024 have their peak correlation with the Warszawa-Okecie synop SMDB (verified 06-06 UTC, dated D) at lag 0.

  • meteorology/monthly (synop): Library: meteorology/api.py stamps pl.datetime(year, month, 1) 00:00 replace_time_zone(‘UTC’); files pruned by year ranges (api.py); request bounds as for daily (request.py, values.py, inclusive). Source: s_m_d_format.txt / s_m_t_format.txt (https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/dane_meteorologiczne/miesieczne/synop/) only say ‘MC Miesiąc’ [month]; Rocznik Meteorologiczny 2024 (https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/Roczniki/Rocznik%20meteorologiczny/Rocznik%20Meteorologiczny%202024.pdf): ‘wyrażone w [°C] i wyznaczone jako średnie arytmetyczne z wartości dobowych’ [monthly Tmax/Tmin means computed as arithmetic means of the daily values]. Checked: Warszawa-Okecie 2024-01..06 s_m_d SUMM/TMAX/TMIN/STM equal sum/max/min/mean of the s_d rows dated in the month, 6/6 months.

  • meteorology/monthly (climate, precipitation): Library: meteorology/api.py stamps pl.datetime(year, month, 1) 00:00 replace_time_zone(‘UTC’); files pruned by year ranges (api.py); request bounds as for daily (request.py, values.py, inclusive). Source: k_m_d_format.txt, k_m_t_format.txt, o_m_format.txt (https://danepubliczne.imgw.pl/data/dane_pomiarowo_obserwacyjne/dane_meteorologiczne/miesieczne/) define ‘MC Miesiąc’ and ‘Miesięczna suma opadów’ without time zone or period bounds; searched Opis.txt, UWAGA!.txt, lista_zmian.txt: silent. Not checked against daily data for these datasets.

ipma#

network

resolution

datasets

source time zone

reference time

verified

observation

hourly

all

unverified (docs silent for observations; live data match UTC)

unverified (docs: hourly mean/sum, stamp position not stated; live data match end of the hour, UTC)

no

Notes:

  • observation/hourly: Library: src/wetterdienst/provider/ipma/observation/parser.py parses the feed key ‘YYYY-mm-ddThh:mi’ with ‘%Y-%m-%dT%H:%M’ and replace_time_zone(‘UTC’); no shift. The whole rolling ~24 h feed is read (api.py); afterwards src/wetterdienst/model/values.py keeps rows with start <= timestamp <= end (closed both), naive start/end tagged and aware ones converted to UTC at src/wetterdienst/model/request.py. Live check 2026-10-10: radiation centroid of 88 mainland stations lies +0.57 h after solar noon (end-of-hour UTC predicts +0.5), and hourly-mean temperature matches METAR means over [T-1h, T] UTC best (LPFR MAE 0.37 K vs 1.04/0.87 K at -1h/+1h; LPPR 0.88 vs 1.17/1.33 K). Source: observations.json section (Portuguese): “YYYY-mm-ddThh:mi: data/hora da observação” [date/time of the observation]; “temperatura: temperatura do ar registada a 1.5 metros de altura, média da hora” [hourly mean]; “precAcumulada: precipitação registada a 1.5 metros de altura, valor acumulado da hora” [hourly total]; “radiacao: radiação solar (kJ/m2)”; “Taxa de atualização horária.” The only UTC sentence on the page is about forecasts: “As previsões são apresentadas em horas UTC. No continente e na Região Autónoma da Madeira a hora legal é igual à UTC, no horário de inverno.” [The forecasts are given in UTC hours …]. https://api.ipma.pt/ (searched the whole page for UTC/hora/fuso).

knmi#

network

resolution

datasets

source time zone

reference time

verified

observation

10_minutes

data

UTC

end (of the preceding 10 minutes; ta/td/rh/p0/pp ‘past minute mean’ at that end; dd/ff partly ‘middle’ at listed stations); the library stamps the end (the file name)

yes

observation

hourly

data

UTC

end (of the preceding hour; T/TD/U/P ‘past 1 minute’ values at that end; FH/FX/RH/DR/SQ/Q hourly aggregates); the library stamps the end (file name + 1 h)

yes

observation

daily

data

UTC

start (UTC day 00-24 labelled with its date; KNMI’s NetCDF time is the end)

yes

Notes:

  • observation/10_minutes (data): Library: the file name is the stamp (_stamp): the file KMDS__OPER_P___10M_OBS_L2_202610140900.nc has the time coordinate 09:00 (long_name ‘Time (in UTC) indicating the end of the 10-minute measurement interval’) and ta 14.7 degC at De Bilt, the temperature NOAA GHCNh De Bilt reads at 09:00 UTC (the files named 08:50 and 09:10 hold 14.6 and 15.2). Read from real files with a key, 2026-10-18; the stamp is computed from the name, the time coordinate is not read. The dataset page’s example table (a file named 2350 with the timestamp 00:00) does not hold for these files. Bounds converted to UTC (src/wetterdienst/model/request.py). Source: “Timestamps in the file and filename are stored in UTC. … The timestamp in the files indicates the end of the measuring interval preceding that timestamp. The filenames use an abbreviated timestamp indicating the total interval.” Example table: ‘10-minute | 2025-01-02T23:50:00Z 2025-01-03T00:00:00Z | KMDS__OPER_P___10M_OBS_L2_202501022350.nc | 2025-01-03T00:00:00Z’ (https://dataplatform.knmi.nl/dataset/docs/10-minute-in-situ-meteorological-observations-1-0; same table on the hourly/daily docs pages); ta: ‘Past minute mean air temperature at 1.50 meters’.

  • observation/hourly (data): Library: the file is named for the START of the hour and its values are stamped with the END, name + 1 h (_stamp, GH-2824; before it the values were stamped one hour early, T, TD, U, P at the start of the hour before their instant). Read from real files with a key, 2026-10-18: hourly-observations-validated-20261014-08.nc has the time coordinate 09:00 (‘Time (in UTC) indicating the end of the hourly measurement interval’) and T 14.7 degC at De Bilt, the value of the 10-minute file named 09:00 and of NOAA GHCNh at 09:00; over the five hours 08 to 12 the mean absolute difference to GHCNh is 0.0 K at the time coordinate and 1.2 K at the name. Key-less export (daggegevens.knmi.nl, De Bilt, June 2024, 718 hours) against GHCNh, hour-ending slot HH read as HH:00: T 0.04 K, TD 0.03 K, P 0.0 hPa, against 0.77 K, 0.58 K, 0.32 hPa one hour off. Source: “Timestamps in the file and filename are stored in UTC. … The timestamp in the files indicates the end of the measuring interval preceding that timestamp. The filenames use an abbreviated timestamp indicating the total interval.” Example table: ‘Hourly | 2025-01-02T23:00:00Z 2025-01-03T00:00:00Z | hourly-observations-20250102-23.nc | 2025-01-03T00:00:00Z’ (https://dataplatform.knmi.nl/dataset/docs/hourly-in-situ-meteorological-observations-validated-1-0; same table on the hourly/daily docs pages); ‘T Past 1 minute air temperature at 1.50 meters’. Legacy key-less service: ‘HH : tijd (HH=uur; UT.12 UT=13 MET; 14 MEZT. Uurvak 05 loopt van 04.00 UT tot 5.00 UT’ (Dutch: hour slot 05 runs from 04.00 UT to 5.00 UT) (https://www.daggegevens.knmi.nl/klimatologie/uurgegevens).

  • observation/daily (data): Library: parser.py stamps the requested date (filename date = start of the UTC day) at 00:00 UTC, ignoring the end-of-day time coordinate; start floored to the day for single-resolution requests (api.py). Source: “Timestamps in the file and filename are stored in UTC. … The timestamp in the files indicates the end of the measuring interval preceding that timestamp. The filenames use an abbreviated timestamp indicating the total interval.” Example table: ‘Daily | 2025-01-02T00:00:00Z 2025-01-03T00:00:00Z | daily-observations-20250102.nc | 2025-01-03T00:00:00Z’ (https://dataplatform.knmi.nl/dataset/docs/daily-in-situ-meteorological-observations-validated-1-0; same table on the hourly/daily docs pages); “Daily mean pressure is calculated as a mean of the 24 pressure values of that day (midnight to midnight UTC).” Read from a real file with a key, 2026-10-18: daily-observations-validated-20261014.nc has the time coordinate 2026-10-15 00:00 (‘the end of the daily measurement interval’) and De Bilt TG 11.5, TN 6.1, TX 17.8, RH 0.0, the values of 20261014 in the legacy export.

lhmt#

network

resolution

datasets

source time zone

reference time

verified

observation

hourly

data

UTC

mixed (instant ‘time of observation’ for temperature/humidity/wind/cloud/pressure/snow; precipitation hourly sum and windGust hourly max: start/end unstated)

partly

Notes:

  • observation/hourly (data): Library: parser.py parses observationTimeUtc (‘%Y-%m-%d %H:%M:%S’) and tags UTC; api.py fetches one file per UTC date in [start, end] after astimezone(UTC); framework trims inclusively. Verified 2025-03-30 (vilniaus-ams): 24 unique UTC stamps 00-23, no gap/duplicate. Source: Lithuanian: ‘observationTimeUtc - atliktų stebėjimų laikas (UTC laiko juosta).’ (time of the observations made, UTC time zone); ‘date - data, kurios stebėjimų duomenis pateikti, YYYY-MM-DD (naudojama UTC laiko juosta).’ (date whose data to return, UTC); ‘windGust - vėjo gūsis, m/s. Maksimalus gūsis per valandą.’ (maximum gust per hour); ‘precipitation - kritulių kiekis, mm. Kritulių suma per valandą.’ (precipitation sum per hour) (https://api.meteo.lt/)

noaa#

network

resolution

datasets

source time zone

reference time

verified

ghcn

hourly

all

UTC

mixed (instant reports, the one nearest to the full hour and stamped with it; precipitation is a running total up to the report, and the closing report of each hour is the one returned)

yes

ghcn

daily

all

local (station’s own observing day; docs give no zone except that TAVG from source ‘S’ runs to 2400 UTC)

unverified (date of the observing day; COOP stations report at an OBS-TIME, e.g. 0700)

partly

Notes:

  • ghcn/hourly: Library: The PSV `DATE` column is parsed with `str.to_datetime(‘%Y-%m-%dT%H:%M:%S’, time_zone=’UTC’)` and not shifted (src/wetterdienst/provider/noaa/ghcn/api.py), then every parameter is thinned to one reading for each full hour, the one nearest to it (a half hour goes to the later hour), stamped with that hour: the METAR of 14:53 is 15:00 (GH-3000, `_one_report_per_hour`). The precipitation is the closing report of its UTC clock hour and is stamped the same way. No window is sent upstream (whole period-of-record file, api.py); the request window is applied afterwards, both ends inclusive (model/values.py). Source: “Each observation is also indexed to a single date-time in ISO 8601 format, plus the Year, Month, Day, Hour and Minute in Coordinated Universal Time (UTC).” and “precipitation = total liquid precipitation (rain or melted snow). Totals are nominally for the hour, but may include intermediate reports within the hour.” and “When multiple observations are transmitted within the hour, the precipitation totals represent a running total for that hour. To extract the total precipitation for a particular hour, the last report should be used.” (https://www.ncei.noaa.gov/oa/global-historical-climatology-network/hourly/doc/ghcnh_DOCUMENTATION.pdf, sections III, IV, IX)

  • ghcn/daily: Library: `DATE` (YYYY-MM-DD) is cast to UTC midnight and not shifted, so a day is stamped at 00:00 UTC of its date whatever the station’s zone (src/wetterdienst/provider/noaa/ghcn/api.py); until GH-2683 the station’s base (standard, DST-free) UTC offset from tzfpy was subtracted, which stamped NYC 05:00 UTC and Tokyo 15:00 UTC the day before. Request bounds without zone are UTC, both ends inclusive (model/request.py, model/values.py), so start=end=D returns D. Source: “TAVG = Average daily temperature (tenths of degrees C) [Note that TAVG from source ‘S’ corresponds to an average of hourly readings for the period ending at 2400 UTC rather than local midnight or other Local Standard Time according to a specific Met Service’s protocol]” (https://www.ncei.noaa.gov/pub/data/ghcn/daily/readme.txt, section III); “OBS-TIME = 4-character time of observation in hour-minute format (i.e. 0700 =7:00 am)” (https://www.ncei.noaa.gov/pub/data/ghcn/daily/by_year/readme-by_year.txt). Searched the readme for UTC/GMT/local/midnight/observation time: no general statement of the day’s zone or of which 24 h a date covers.

rmi#

network

resolution

datasets

source time zone

reference time

verified

observation

10_minutes

all

UTC

unverified (observed: end; the record opens at 00:10, the 10-minute record’s temporal coverage begins 1995-01-01T00:10:00)

partly

observation

hourly

all

UTC

unverified (observed: end, the hour before the stamp)

partly

observation

daily

all

UTC

start (of the UTC day 00:00-24:00)

partly

Notes:

  • observation/10_minutes: Library: Bounds converted to UTC and sent as CQL ‘timestamp >= <start>Z AND timestamp <= <end>Z’ (inclusive; ‘DURING X/X’ is a 400) (rmi/observation/api.py); ‘Z’ stamps parsed and labelled UTC unchanged (api.py); output cut inclusively (model/values.py). Source: No stamp/zone sentence in the 10-min record (https://opendata.meteo.be/geonetwork/srv/api/records/RMI_DATASET_AWS_10MIN/formatters/xml): only “These weather stations report meteorological paramaters … every 10 minutes.” and the temporal extent “1995-01-01T00:10:00” to “now” (read 2026-10-18), which opens at 00:10, as a series of intervals ending at the stamp does The WFS stamps end in ‘Z’ (e.g. ‘2024-09-30T00:10:00Z’). Daily record: “Daily AWS data are computed from the RMI 10-min AWS observations available during the considered day from 00:00 to 24:00 UTC.” (https://opendata.meteo.be/geonetwork/srv/api/records/RMI_DATASET_AWS_1DAY/formatters/xml). Data check (not documentation), Uccle 6447 Oct 2024: daily precip equals the sum of 10-min values stamped (D 00:00, D+1 00:00] (2024-10-09: 24.94 vs 24.91 for [D 00:00, D+1 00:00); 2024-10-10: 6.19 vs 6.22), i.e. the 10-min stamp is the end of its interval.

  • observation/hourly: Library: Bounds converted to UTC and sent as CQL ‘timestamp >= <start>Z AND timestamp <= <end>Z’ (inclusive; ‘DURING X/X’ is a 400) (rmi/observation/api.py); ‘Z’ stamps parsed and labelled UTC unchanged (api.py); output cut inclusively (model/values.py). Source: “Hourly AWS data are computed from the RMI 10-min AWS observations available during the 1-hour period. The hourly data correspond to the sum of the 10-min observations in case of cumulative parameters (e.g., precipitation quantity, sunshine duration) and otherwise to the average of the 10-min observations.” https://opendata.meteo.be/geonetwork/srv/api/records/RMI_DATASET_AWS_1HOUR/formatters/xml (no stamp position, no zone; WFS stamps end in ‘Z’). Data check (not documentation), 6447 Sep 30-Oct 15 2024: hourly precip at H = sum of 10-min values stamped H-50..H in 58/58 wet hours ((H, H+1h]: 8; [H, H+1h): 11).

  • observation/daily: Library: Bounds converted to UTC and sent as CQL ‘timestamp >= <start>Z AND timestamp <= <end>Z’ (inclusive; ‘DURING X/X’ is a 400) (rmi/observation/api.py); ‘Z’ stamps parsed and labelled UTC unchanged (api.py); output cut inclusively (model/values.py). Source: “Daily AWS data are computed from the RMI 10-min AWS observations available during the considered day from 00:00 to 24:00 UTC. The daily data correspond to the sum of the 10-min observations in case of cumulative parameters (e.g., precipitation quantity, sunshine duration) and otherwise to the average, minimum of maximum of the 10-min observations.” https://opendata.meteo.be/geonetwork/srv/api/records/RMI_DATASET_AWS_1DAY/formatters/xml ; daily stamps are ‘D T00:00:00Z’ (stamp position inferred from the day naming, not stated).

wsv#

network

resolution

datasets

source time zone

reference time

verified

pegel

1_minute

all

Europe/Berlin local legal time (CET/CEST) with explicit UTC offset in each value

unverified (raw measurements; docs do not say instant or mean)

partly

pegel

5_minutes

all

Europe/Berlin local legal time (CET/CEST) with explicit UTC offset in each value

unverified (raw measurements; docs do not say instant or mean)

partly

pegel

10_minutes

all

Europe/Berlin local legal time (CET/CEST) with explicit UTC offset in each value

unverified (raw measurements; docs do not say instant or mean)

partly

pegel

15_minutes

all

Europe/Berlin local legal time (CET/CEST) with explicit UTC offset in each value

unverified (raw measurements; docs do not say instant or mean)

partly

pegel

hourly

all

Europe/Berlin local legal time (CET/CEST) with explicit UTC offset in each value

unverified (raw measurements; docs do not say instant or mean)

partly

Notes:

  • pegel/1_minute: Library: `timestamp` parsed with `%Y-%m-%dT%H:%M:%S%z` into UTC, so the offset decides and the repeated autumn hour stays distinct (src/wetterdienst/provider/wsv/pegel/api.py). The measurements URL carries no start/end (api.py), so upstream returns its default of the last 10 days, not the 31 it holds; the request window is applied afterwards, both ends inclusive. Source: “timestamp Alle Zeitangaben erfolgen in lokaler gesetzlicher Zeit (MEZ/MESZ mit Sommerzeitumstellung): JSON-Format: Zeitstempel werden vollqualifiziert nach ISO 8601 mit explizitem Zeitzonenoffset zu UTC kodiert. Beispiel: 2025-09-25T10:00:00+02:00” (German: all times are in local legal time (CET/CEST with DST change); JSON: timestamps fully qualified ISO 8601 with explicit offset to UTC) and “Wasserstandsrohdaten des Pegels Bonn der letzten 10 Tage (Standardzeitraum).” (raw water-level data of the last 10 days (default period)) (https://www.pegelonline.wsv.de/webservice/dokuRestapi). Searched for Momentan/Mittelwert/Rohdaten: no statement of instant vs. mean.

  • pegel/5_minutes: Library: `timestamp` parsed with `%Y-%m-%dT%H:%M:%S%z` into UTC, so the offset decides and the repeated autumn hour stays distinct (src/wetterdienst/provider/wsv/pegel/api.py). The measurements URL carries no start/end (api.py), so upstream returns its default of the last 10 days, not the 31 it holds; the request window is applied afterwards, both ends inclusive. Source: “timestamp Alle Zeitangaben erfolgen in lokaler gesetzlicher Zeit (MEZ/MESZ mit Sommerzeitumstellung): JSON-Format: Zeitstempel werden vollqualifiziert nach ISO 8601 mit explizitem Zeitzonenoffset zu UTC kodiert. Beispiel: 2025-09-25T10:00:00+02:00” (German: all times are in local legal time (CET/CEST with DST change); JSON: timestamps fully qualified ISO 8601 with explicit offset to UTC) and “Wasserstandsrohdaten des Pegels Bonn der letzten 10 Tage (Standardzeitraum).” (raw water-level data of the last 10 days (default period)) (https://www.pegelonline.wsv.de/webservice/dokuRestapi). Searched for Momentan/Mittelwert/Rohdaten: no statement of instant vs. mean.

  • pegel/10_minutes: Library: `timestamp` parsed with `%Y-%m-%dT%H:%M:%S%z` into UTC, so the offset decides and the repeated autumn hour stays distinct (src/wetterdienst/provider/wsv/pegel/api.py). The measurements URL carries no start/end (api.py), so upstream returns its default of the last 10 days, not the 31 it holds; the request window is applied afterwards, both ends inclusive. Source: “timestamp Alle Zeitangaben erfolgen in lokaler gesetzlicher Zeit (MEZ/MESZ mit Sommerzeitumstellung): JSON-Format: Zeitstempel werden vollqualifiziert nach ISO 8601 mit explizitem Zeitzonenoffset zu UTC kodiert. Beispiel: 2025-09-25T10:00:00+02:00” (German: all times are in local legal time (CET/CEST with DST change); JSON: timestamps fully qualified ISO 8601 with explicit offset to UTC) and “Wasserstandsrohdaten des Pegels Bonn der letzten 10 Tage (Standardzeitraum).” (raw water-level data of the last 10 days (default period)) (https://www.pegelonline.wsv.de/webservice/dokuRestapi). Searched for Momentan/Mittelwert/Rohdaten: no statement of instant vs. mean.

  • pegel/15_minutes: Library: `timestamp` parsed with `%Y-%m-%dT%H:%M:%S%z` into UTC, so the offset decides and the repeated autumn hour stays distinct (src/wetterdienst/provider/wsv/pegel/api.py). The measurements URL carries no start/end (api.py), so upstream returns its default of the last 10 days, not the 31 it holds; the request window is applied afterwards, both ends inclusive. Source: “timestamp Alle Zeitangaben erfolgen in lokaler gesetzlicher Zeit (MEZ/MESZ mit Sommerzeitumstellung): JSON-Format: Zeitstempel werden vollqualifiziert nach ISO 8601 mit explizitem Zeitzonenoffset zu UTC kodiert. Beispiel: 2025-09-25T10:00:00+02:00” (German: all times are in local legal time (CET/CEST with DST change); JSON: timestamps fully qualified ISO 8601 with explicit offset to UTC) and “Wasserstandsrohdaten des Pegels Bonn der letzten 10 Tage (Standardzeitraum).” (raw water-level data of the last 10 days (default period)) (https://www.pegelonline.wsv.de/webservice/dokuRestapi). Searched for Momentan/Mittelwert/Rohdaten: no statement of instant vs. mean.

  • pegel/hourly: Library: `timestamp` parsed with `%Y-%m-%dT%H:%M:%S%z` into UTC, so the offset decides and the repeated autumn hour stays distinct (src/wetterdienst/provider/wsv/pegel/api.py). The measurements URL carries no start/end (api.py), so upstream returns its default of the last 10 days, not the 31 it holds; the request window is applied afterwards, both ends inclusive. Source: “timestamp Alle Zeitangaben erfolgen in lokaler gesetzlicher Zeit (MEZ/MESZ mit Sommerzeitumstellung): JSON-Format: Zeitstempel werden vollqualifiziert nach ISO 8601 mit explizitem Zeitzonenoffset zu UTC kodiert. Beispiel: 2025-09-25T10:00:00+02:00” (German: all times are in local legal time (CET/CEST with DST change); JSON: timestamps fully qualified ISO 8601 with explicit offset to UTC) and “Wasserstandsrohdaten des Pegels Bonn der letzten 10 Tage (Standardzeitraum).” (raw water-level data of the last 10 days (default period)) (https://www.pegelonline.wsv.de/webservice/dokuRestapi). Searched for Momentan/Mittelwert/Rohdaten: no statement of instant vs. mean.

ea#

network

resolution

datasets

source time zone

reference time

verified

hydrology

15_minutes

all

unverified

instant

partly

hydrology

daily

all

unverified

unverified (data: start, the 24 h from 09:00 D to 09:00 D+1, stamped D 00:00 UTC; the mean is dated D 09:00, the maximum and minimum at the instant of the extreme)

no

Notes:

  • hydrology/15_minutes: Library: Zone-less `dateTime` parsed as UTC (src/wetterdienst/provider/ea/hydrology/api.py); request bounds sent as zone-less `mineq-dateTime`/`maxeq-dateTime` with a 1-day margin each side (api.py), window applied afterwards, both ends inclusive. Source: Table of the reference: “Flow 15min instantaneous m3/s” / “Level 15min instantaneous m”; “dateTime The date and time of the reading xsd:dateTime”; “When filtering on a range of date-times, to ensure accurate comparison, you should include the timezone on the timestamp that you want to compare values to.” (https://environment.data.gov.uk/hydrology/doc/reference). Searched for UTC/GMT/BST/time zone: the zone of returned values is not stated. Data: 96 readings with no gap or duplicate on 2025-03-30 and 2025-10-26 (Felsted), and the 2026-10-01 (BST) levels match the flood-monitoring API’s ‘…Z’ stamps exactly at +0 h (96/96), so the values are UTC.

  • hydrology/daily: Library: Zone-less `dateTime` parsed as UTC; the day D is taken from the stamp 9 hours earlier and the value is stamped 00:00 UTC of D (src/wetterdienst/provider/ea/hydrology/api.py). The mean is dated D 09:00:00, the maximum and minimum at the instant of the extreme inside the 24 h from D 09:00 to D+1 09:00 (flow-max-86400, station 052d0819-2a32-47df-9b99-c243c9c8235b: 2025-06-04T03:30 is the maximum of 2025-06-03, 2025-06-05T07:00 that of 2025-06-04, 2025-06-05T13:30 that of 2025-06-05), so all of them are filed under D (the `date` field of a reading is the calendar date of its instant and is not used). Where the endpoint answers a day twice, with an unchecked or incomplete reading or a null beside the complete one, the one with a value is kept, then the complete one, then the one of quality Good, and of equal standing the later one. Until GH-2683 the stamps were kept as dated, 09:00 UTC for the mean. Source: “Flow daily mean, min, max m3/s” / “Level daily min, max m” and “date The date of the reading” (https://environment.data.gov.uk/hydrology/doc/reference); no statement of the daily window or zone found. Data (Felsted flow-m-86400): the value stamped D 09:00 equals the mean of the 15-min readings in [D 09:00, D+1 09:00) (mean abs error 0.0002 m3/s in Jan 2025, 0.0003 in Jul 2024, against 0.025 for windows shifted by 1 h).

nws#

network

resolution

datasets

source time zone

reference time

verified

observation

hourly

all

UTC offset carried in each value (ISO 8601 date-time, observed ‘+00:00’)

mixed (instant report; *LastHour/Last3Hours/Last6Hours fields cover the period ending at the stamp by their names)

partly

Notes:

  • observation/hourly: Library: `timestamp` parsed with `%Y-%m-%dT%H:%M:%S%z` into UTC (src/wetterdienst/provider/nws/observation/api.py); request start/end sent upstream converted to UTC with `astimezone(utc)` and widened by 30 minutes on both sides (api.py), window applied again afterwards, both ends inclusive. Of the endpoint’s observations (a 5-minute ASOS row at each multiple of five minutes, the METAR at :51 to :54, SPECI) one is kept for each full hour (GH-2857): a METAR or SPECI over a 5-minute row, then the one nearest to the full hour it is nearest to, then the earlier; it is stamped with that full hour, not with its own minute (the METAR of 14:54 is 15:00). Source: OpenAPI Observation schema: “timestamp”: {“type”: “string”, “format”: “date-time”}; property names “maxTemperatureLast24Hours”, “precipitationLastHour”, “precipitationLast3Hours”, “precipitationLast6Hours” (https://api.weather.gov/openapi.json); a live value reads ‘2026-10-10T19:45:00+00:00’ (KJFK). No prose definition of the reference time found in the spec.

eaufrance#

network

resolution

datasets

source time zone

reference time

verified

hubeau

5_minutes

all

UTC

instant

yes

hubeau

6_minutes

all

UTC

instant

yes

hubeau

10_minutes

all

UTC

instant

yes

hubeau

15_minutes

all

UTC

instant

yes

hubeau

hourly

all

UTC

instant

yes

Notes:

  • hubeau/5_minutes: Library: `date_obs` (’…Z’) parsed with `%Y-%m-%dT%H:%M:%SZ` and labelled UTC (src/wetterdienst/provider/eaufrance/hubeau/api.py). The last 30 days are always fetched in UTC chunks regardless of the request (api.py); the request window is applied afterwards, both ends inclusive. Source: “date_obs”: “Date d’observation hydro : date, exprimée en TU, d’occurence de la donnée.” (French: hydro observation date: date, expressed in UT, of the occurrence of the datum) and date_debut_obs/date_fin_obs “(exprimée en TU)” (https://hubeau.eaufrance.fr/api/v2/hydrometrie/api-docs, observations_tr)

  • hubeau/6_minutes: Library: `date_obs` (’…Z’) parsed with `%Y-%m-%dT%H:%M:%SZ` and labelled UTC (src/wetterdienst/provider/eaufrance/hubeau/api.py). The last 30 days are always fetched in UTC chunks regardless of the request (api.py); the request window is applied afterwards, both ends inclusive. Source: “date_obs”: “Date d’observation hydro : date, exprimée en TU, d’occurence de la donnée.” (French: hydro observation date: date, expressed in UT, of the occurrence of the datum) and date_debut_obs/date_fin_obs “(exprimée en TU)” (https://hubeau.eaufrance.fr/api/v2/hydrometrie/api-docs, observations_tr)

  • hubeau/10_minutes: Library: `date_obs` (’…Z’) parsed with `%Y-%m-%dT%H:%M:%SZ` and labelled UTC (src/wetterdienst/provider/eaufrance/hubeau/api.py). The last 30 days are always fetched in UTC chunks regardless of the request (api.py); the request window is applied afterwards, both ends inclusive. Source: “date_obs”: “Date d’observation hydro : date, exprimée en TU, d’occurence de la donnée.” (French: hydro observation date: date, expressed in UT, of the occurrence of the datum) and date_debut_obs/date_fin_obs “(exprimée en TU)” (https://hubeau.eaufrance.fr/api/v2/hydrometrie/api-docs, observations_tr)

  • hubeau/15_minutes: Library: `date_obs` (’…Z’) parsed with `%Y-%m-%dT%H:%M:%SZ` and labelled UTC (src/wetterdienst/provider/eaufrance/hubeau/api.py). The last 30 days are always fetched in UTC chunks regardless of the request (api.py); the request window is applied afterwards, both ends inclusive. Source: “date_obs”: “Date d’observation hydro : date, exprimée en TU, d’occurence de la donnée.” (French: hydro observation date: date, expressed in UT, of the occurrence of the datum) and date_debut_obs/date_fin_obs “(exprimée en TU)” (https://hubeau.eaufrance.fr/api/v2/hydrometrie/api-docs, observations_tr)

  • hubeau/hourly: Library: `date_obs` (’…Z’) parsed with `%Y-%m-%dT%H:%M:%SZ` and labelled UTC (src/wetterdienst/provider/eaufrance/hubeau/api.py). The last 30 days are always fetched in UTC chunks regardless of the request (api.py); the request window is applied afterwards, both ends inclusive. Source: “date_obs”: “Date d’observation hydro : date, exprimée en TU, d’occurence de la donnée.” (French: hydro observation date: date, expressed in UT, of the occurrence of the datum) and date_debut_obs/date_fin_obs “(exprimée en TU)” (https://hubeau.eaufrance.fr/api/v2/hydrometrie/api-docs, observations_tr)

geosphere#

network

resolution

datasets

source time zone

reference time

verified

observation

10_minutes

all

UTC

unverified (observed: end, the 10 minutes before the stamp)

partly

observation

hourly

all

UTC

unverified (observed: end, the hour before the stamp; tl the value at the stamp)

partly

observation

daily

all

UTC

mixed (stamp D 00:00 UTC names day D; rr D 06:01->D+1 06:00 UTC, tlmax/tlmin D-1 18:10->D 18:00 UTC, cglo_j/so_h/ffx D-1 23->D 23 UTC (0-24 MOZ), p/rf/vv/bewm terms 06,13,18 UTC, sh at 06 UTC)

yes

observation

monthly

all

UTC

unverified (observed: start)

partly

Notes:

  • observation/10_minutes: Library: API ‘timestamps’ (‘2024-10-01T00:00+00:00’) parsed with ‘%Y-%m-%dT%H:%M+%Z’ and labelled UTC unchanged (geosphere/observation/api.py); request bounds converted to UTC, widened by 1 day each side and sent as UTC (api.py), then cut inclusively (model/values.py). Source: “Zeitzone: UTC” (dataset page, German) [time zone: UTC] https://data.hub.geosphere.at/en/dataset/klima-v2-10min?lang=de ; parameter descriptions (https://dataset.api.hub.geosphere.at/v1/station/historical/klima-v2-10min/metadata) say e.g. “Lufttemperatur in 2m Höhe, 10-Minuten-Messwert”, “Sonnenscheindauer in Sekunden, 10-Minutensumme”, no stamp position. The 1d metadata describes the daily max as “Maximum der Zehnminutenwerte (tlmax) von 18:10 UTC Vortag - 18:00 UTC aktueller Tag” and rr as “06:01 UTC bis 06:00 UTC des Folgetags”, which reads as 10-min values labelled by their end. Data check (not documentation), station 5904, Oct 2024: daily rr = sum of 10-min rr stamped (D 06:00, D+1 06:00] in 10/10 days ([D 06:00, D+1 06:00): 6/10). Searched also DataHub-DE-Informationen-zu-Datenreihen.pdf and -zur-Datenqualität.pdf: no time statement.

  • observation/hourly: Library: API ‘timestamps’ (‘2024-10-01T00:00+00:00’) parsed with ‘%Y-%m-%dT%H:%M+%Z’ and labelled UTC unchanged (geosphere/observation/api.py); request bounds converted to UTC, widened by 1 day each side and sent as UTC (api.py), then cut inclusively (model/values.py). Source: “Zeitzone: UTC” (dataset page, German) [time zone: UTC] https://data.hub.geosphere.at/dataset/klima-v2-1h ; descriptions e.g. “Niederschlag, Stundensumme”, “Lufttemperatur in 2m Höhe, Stunden-Messwert” (klima-v2-1h/metadata) - no stamp position. Data check (not documentation), 5904 Oct 2024: hourly rr at H = sum of 10-min rr stamped H-50..H in 56/56 wet hours ([H, H+1h): 7); hourly tl at H = 10-min tl at H (385/385). Hourly tl of Wien Hohe Warte 105 against NOAA GHCNh Wien/Hohe Warte, 448 hours of 2025-06-01 to 20: mean absolute difference 0.23 K at the stamp, 1.10 K and 0.82 K one hour before and after (the hourly mean ends at the stamp, so it is not the instant).

  • observation/daily: Library: API ‘timestamps’ (‘2024-10-01T00:00+00:00’) parsed with ‘%Y-%m-%dT%H:%M+%Z’ and labelled UTC unchanged (geosphere/observation/api.py); request bounds converted to UTC, widened by 1 day each side and sent as UTC (api.py), then cut inclusively (model/values.py). The D 00:00 label is kept; no window is rebuilt. Source: “Zeitzone: UTC” (dataset page, German) [time zone: UTC] https://data.hub.geosphere.at/en/dataset/klima-v2-1d?lang=de ; parameter descriptions (German) from https://dataset.api.hub.geosphere.at/v1/station/historical/klima-v2-1d/metadata : rr “Niederschlag, 24-Stundensumme der Einminutenwerte von 06:01 UTC bis 06:00 UTC des Folgetags” [from 06:01 UTC to 06:00 UTC of the following day]; tlmax “Lufttemperatur in 2m Höhe, Maximum der Zehnminutenwerte (tlmax) von 18:10 UTC Vortag - 18:00 UTC aktueller Tag” [18:10 UTC previous day - 18:00 UTC current day]; cglo_j “kalibrierte 24-Stundensumme aus den Stundenwerten 0-24 Uhr MOZ (23 Vortag - 23 Tag UTC)”; p_mittel “Tagesmittel aus den Beobachtungsterminen I,II,III (06,13,18 UTC)”; tl_mittel “Mittelwert berechnet aus (tlmax + tlmin)/2”. Data check: rr matches the 10-min sums over (D 06:00, D+1 06:00] 10/10, tlmax/tlmin the 10-min extremes over (D-1 18:00, D 18:00] 15/15 (UTC day 13/15, 8/15).

  • observation/monthly: Library: API ‘timestamps’ (‘2024-10-01T00:00+00:00’) parsed with ‘%Y-%m-%dT%H:%M+%Z’ and labelled UTC unchanged (geosphere/observation/api.py); request bounds converted to UTC, widened by 1 day each side and sent as UTC (api.py), then cut inclusively (model/values.py). Source: “Zeitzone: UTC” (dataset page, German) [time zone: UTC] https://data.hub.geosphere.at/en/dataset/klima-v2-1m?lang=de ; rr “Niederschlag, Monatssumme aus 24-Stundensummen zum Terminwert I (6 UTC)” (klima-v2-1m/metadata). Docs silent on the stamp position; data check (not documentation): monthly rr stamped 2024-09-01 = 242.0 mm = sum of the September daily rr (242.4), 2024-10-01 = 44.0 = October (44.1), so the stamp is the first day of the month it describes.

meteofrance#

network

resolution

datasets

source time zone

reference time

verified

synop

subdaily

data

UTC

mixed (instant for t/td/u/vv/n/pres/pmer; wind dd/ff: a 10-min mean, its position relative to the stamp not stated)

partly

synop

subdaily

data

UTC

end (of the preceding 1/3/6/12/24 h for rr1..rr24, 10 min for raf10, 24 h for tn24/tx24)

partly

observation

daily

core

UTC for metropolitan France; ‘FU’ for overseas departments 971-975/984-988, a fixed offset per territory without DST (measured against SYNOP, see the hourly note): the date is the FU date and is kept

start (date label J; RR window 06 FU J -> 06 FU J+1, i.e. 06 UTC in metropole)

yes

observation

daily

core

UTC for metropolitan France; ‘FU’ for overseas departments 971-975/984-988, a fixed offset per territory without DST (measured against SYNOP, see the hourly note): the date is the FU date and is kept

unverified (observed: start (date label J; TX window 06 UTC J -> 06 UTC J+1))

partly

observation

daily

core

UTC for metropolitan France; ‘FU’ for overseas departments 971-975/984-988, a fixed offset per territory without DST (measured against SYNOP, see the hourly note): the date is the FU date and is kept

unverified (observed: mixed (date label J; TN window 18 UTC J-1 -> 18 UTC J))

partly

observation

daily

core

UTC for metropolitan France; ‘FU’ for overseas departments 971-975/984-988, a fixed offset per territory without DST (measured against SYNOP, see the hourly note): the date is the FU date and is kept

unverified (observed: start (calendar day J 00-24 UTC for TM; FFM/FXI/DXI window unverified))

partly

observation

daily

others

UTC for metropolitan France; ‘FU’ for overseas departments 971-975/984-988, a fixed offset per territory without DST (measured against SYNOP, see the hourly note): the date is the FU date and is kept

mixed (date label J; HNEIGEF 06 FU J -> 06 FU J+1; NEIGETOT06 instant at 06 h; PMERM/TSVM/INST/GLOT/UM window unverified)

partly

observation

6_minutes

data

UTC for metropolitan France; ‘FU’ for overseas departments 971-975/984-988, a fixed offset per territory without DST (measured against SYNOP, see the hourly note): converted to UTC by the library

unverified (observed: end (of the preceding 6 minutes))

partly

observation

hourly

core, others

UTC for metropolitan France; ‘FU’ for overseas departments 971-975/984-988, a fixed offset per territory without DST (measured against SYNOP, see the hourly note): converted to UTC by the library

mixed (instant for T/TD/U/PMER/PSTAT/N/VV; wind FF/DD: a 10-min mean, its position relative to the stamp not stated)

partly

observation

hourly

core

UTC for metropolitan France; ‘FU’ for overseas departments 971-975/984-988, a fixed offset per territory without DST (measured against SYNOP, see the hourly note): converted to UTC by the library

unverified (observed: end (of the preceding hour: RR1, TN, TX, FXY))

partly

observation

hourly

others

UTC for metropolitan France; ‘FU’ for overseas departments 971-975/984-988, a fixed offset per territory without DST (measured against SYNOP, see the hourly note): converted to UTC by the library, GLO/INS too although the source calls them ‘en heure UTC’ (TSV variants GLO2/INS2 not read; see the note)

unverified (GLO, INS hourly sums: start or end of the hour not stated)

partly

observation

monthly

data

UTC for metropolitan France; ‘FU’ for overseas departments 971-975/984-988, a fixed offset per territory without DST (measured against SYNOP, see the hourly note): the date is the FU date and is kept

start (calendar month label; aggregates of the daily values of days 1..last, so RR runs 06 UTC on the 1st -> 06 UTC on the 1st of the next month)

partly

Notes:

  • synop/subdaily (data): Library: synop/api.py parses validity_time with ‘%Y-%m-%dT%H:%M:%SZ’ and replace_time_zone(‘UTC’) (no shift); yearly files picked by UTC year of start/end (synop/api.py); bounds naive->UTC, closed both (model/request.py, model/values.py). Source: “validity_time | date et heure de la mesure réalisée par le capteur | car | ISO 8601/UTC AAAA-MM-DDTHH:MM:SSZ” (French) = “date and time of the measurement made by the sensor, ISO 8601/UTC”; “t température”, “dd direction du vent moyen 10 min”, “ff vitesse du vent moyen 10 min” (French) = “direction of the mean wind over 10 min”, “speed of the mean wind over 10 min”: which end of the 10 minutes the stamp is is not stated (only raf10 says “sur les 10 dernières minutes”). https://static.data.gouv.fr/resources/archive-synop-omm/20260422-141723/observations-descriptif-technique-donnees-synop-omm-20260420.pdf . Cross-check: hourly T/TD/U/PMER/PSTAT/FF/DD of Paris-Montsouris 75114001 equal SYNOP 07156 at lag 0 (mean|diff| 0.0-0.03).

  • synop/subdaily (data): Library: Same as above: validity_time parsed as UTC, no shift (synop/api.py). Source: “rr1 précipitations sur la dernière heure”, “rr3 précipitations dans les 3 dernières heures” … “rr24 précipitations dans les 24 dernières heures”, “raf10 rafales sur les 10 dernières minutes”, “tn24 température minimale sur 24 heures”, “tx24 température maximale sur 24 heures” (French) = “precipitation over the last hour / last 3 … 24 hours”, “gusts over the last 10 minutes”, “min/max temperature over 24 hours” (https://static.data.gouv.fr/resources/archive-synop-omm/20260422-141723/observations-descriptif-technique-donnees-synop-omm-20260420.pdf). The PDF does not say tn24/tx24 end at the stamp; established by comparison: Embrun 07591 tn24/tx24/rr24 equal min/max/sum of hourly TN/TX/RR1 of 05046001 over (t-24h, t] in 447/447 cases (Jun-Jul 2025), 352/447 with a 3 h offset.

  • observation/daily (core): Library: AAAAMMJJ parsed ‘%Y%m%d’ as J 00:00 UTC (observation/api.py); the 06 h offset of the window is not represented: end=J includes the value whose RR runs to J+1 06 UTC; observation/api.py parses the date column with a naive strptime and labels it 00:00 UTC for every department incl. 97x/98x; the FU date of an overseas station is kept, as a daily or monthly stamp names its day or month (GH-2677); request bounds: naive start/end -> UTC (model/request.py), filter is_between closed=’both’ (model/values.py); period buckets picked by calendar year of start/end (observation/api.py). Source: “Les heures sont exprimées en UTC pour la métropole et en FU pour l’outre-mer” (French; dataset description on data.gouv.fr, identical for the 6-minute, hourly, daily and monthly datasets) = “Hours are expressed in UTC for metropolitan France and in FU [local time zone, “fuseau”] for overseas territories”. FU is not defined further in any file fetched. “RR : quantité de précipitation tombée en 24 heures (de 06h FU le jour J à 06h FU le jour J+1). La valeur relevée à J+1 est affectée au jour J” (French) = “precipitation fallen in 24 hours (from 06 FU on day J to 06 FU on day J+1). The value read on J+1 is assigned to day J” https://meteofrance.s3.sbg.io.cloud.ovh.net/data/synchro_ftp/BASE/QUOT/Q_descriptif_champs_RR-T-Vent.csv ; https://www.data.gouv.fr/datasets/donnees-climatologiques-de-base-quotidiennes (dataset 6569b51ae64326786e4e8e1a). Comparison: daily RR == sum of hourly RR1 stamped (J 06, J+1 06] UTC 217/217 (Paris); Réunion matches 06-06 in its own FU stamps 195/217. Overseas, the daily file is on the hourly file’s FU clock: RR equals the sum of the hourly RR1 stamped (J 06, J+1 06] in the file’s own stamps, not (J 00, J+1 00], in every department (e.g. 97301001: 369 of 370 wet days against 62; 98403001: 105 of 105 against 26), so the date J is the FU date.

  • observation/daily (core): Library: Stamped J 00:00 UTC (observation/api.py); observation/api.py parses the date column with a naive strptime and labels it 00:00 UTC for every department incl. 97x/98x; the FU date of an overseas station is kept, as a daily or monthly stamp names its day or month (GH-2677); request bounds: naive start/end -> UTC (model/request.py), filter is_between closed=’both’ (model/values.py); period buckets picked by calendar year of start/end (observation/api.py). Source: “Les heures sont exprimées en UTC pour la métropole et en FU pour l’outre-mer” (French; dataset description on data.gouv.fr, identical for the 6-minute, hourly, daily and monthly datasets) = “Hours are expressed in UTC for metropolitan France and in FU [local time zone, “fuseau”] for overseas territories”. FU is not defined further in any file fetched. “TX : température maximale sous abri (en °C et 1/10)”, “HTX : heure de TX (hhmm)” (Q_descriptif_champs_RR-T-Vent.csv and climatologie-donnees-quotidiennes-descriptif-20260603.pdf): no window stated. Comparison (Paris 75114001, Apr-Oct 2025): TX == max hourly TX over (J 06, J+1 06] 215/215, over 00-24 212/216, over 18-18 175/214. (https://www.data.gouv.fr/datasets/6569b51ae64326786e4e8e1a/)

  • observation/daily (core): Library: Stamped J 00:00 UTC although the TN window starts 6 h before and ends 18 h after the stamp (observation/api.py); observation/api.py parses the date column with a naive strptime and labels it 00:00 UTC for every department incl. 97x/98x; the FU date of an overseas station is kept, as a daily or monthly stamp names its day or month (GH-2677); request bounds: naive start/end -> UTC (model/request.py), filter is_between closed=’both’ (model/values.py); period buckets picked by calendar year of start/end (observation/api.py). Source: “Les heures sont exprimées en UTC pour la métropole et en FU pour l’outre-mer” (French; dataset description on data.gouv.fr, identical for the 6-minute, hourly, daily and monthly datasets) = “Hours are expressed in UTC for metropolitan France and in FU [local time zone, “fuseau”] for overseas territories”. FU is not defined further in any file fetched. “TN : température minimale sous abri (en °C et 1/10)” (Q_descriptif_champs_RR-T-Vent.csv): no window stated. Comparison (Paris): TN == min hourly TN over (J-1 18, J 18] 214/214, over 00-24 175/216, over 06-06 24/215; Réunion: 18-18 in FU stamps 215/217. (https://www.data.gouv.fr/datasets/6569b51ae64326786e4e8e1a/)

  • observation/daily (core): Library: Stamped J 00:00 UTC (observation/api.py); observation/api.py parses the date column with a naive strptime and labels it 00:00 UTC for every department incl. 97x/98x; the FU date of an overseas station is kept, as a daily or monthly stamp names its day or month (GH-2677); request bounds: naive start/end -> UTC (model/request.py), filter is_between closed=’both’ (model/values.py); period buckets picked by calendar year of start/end (observation/api.py). Source: “Les heures sont exprimées en UTC pour la métropole et en FU pour l’outre-mer” (French; dataset description on data.gouv.fr, identical for the 6-minute, hourly, daily and monthly datasets) = “Hours are expressed in UTC for metropolitan France and in FU [local time zone, “fuseau”] for overseas territories”. FU is not defined further in any file fetched. “TM : moyenne quotidienne des températures horaires sous abri”, “FFM : moyenne quotidienne de la force du vent moyenné sur 10 mn, à 10 m”, “FXI : maximum quotidien de la force maximale horaire du vent instantané, à 10 m” (French) = “daily mean of hourly screen temperatures”, “daily mean of 10-min wind”, “daily max of hourly max instantaneous wind” (Q_descriptif_champs_RR-T-Vent.csv). Day not defined. Comparison: TM == mean of hourly T stamped J 01..J+1 00 UTC 208/217 (vs 81/217 for 00..23); FFM/FXI not compared. (https://www.data.gouv.fr/datasets/6569b51ae64326786e4e8e1a/)

  • observation/daily (others): Library: Stamped J 00:00 UTC (observation/api.py); the 06 h reading of NEIGETOT06 and the 06-06 window of HNEIGEF are not represented; observation/api.py parses the date column with a naive strptime and labels it 00:00 UTC for every department incl. 97x/98x; the FU date of an overseas station is kept, as a daily or monthly stamp names its day or month (GH-2677); request bounds: naive start/end -> UTC (model/request.py), filter is_between closed=’both’ (model/values.py); period buckets picked by calendar year of start/end (observation/api.py). Source: “Les heures sont exprimées en UTC pour la métropole et en FU pour l’outre-mer” (French; dataset description on data.gouv.fr, identical for the 6-minute, hourly, daily and monthly datasets) = “Hours are expressed in UTC for metropolitan France and in FU [local time zone, “fuseau”] for overseas territories”. FU is not defined further in any file fetched. “HNEIGEF : hauteur de neige fraîche tombée en 24 heures (de 06h FU le jour J à 06h FU le jour J+1) qui reste au sol à 06h FU. La valeur relevée à J+1 est affectée au jour J”, “NEIGETOT06 : épaisseur totale de neige au sol mesurée à 6h (NEIGETOT de 6h)”, “PMERM : moyenne quotidienne des pressions mer horaires”, “INST : durée d’insolation quotidienne” (French) = “fresh snow fallen in 24 h (06 FU day J to 06 FU day J+1) remaining at 06 FU; value read on J+1 assigned to J”, “total snow depth measured at 6h”, “daily mean of hourly sea-level pressures”, “daily sunshine duration” https://meteofrance.s3.sbg.io.cloud.ovh.net/data/synchro_ftp/BASE/QUOT/Q_descriptif_champs_autres-parametres.csv ; https://www.data.gouv.fr/datasets/donnees-climatologiques-de-base-quotidiennes (dataset 6569b51ae64326786e4e8e1a). Not compared against a second source.

  • observation/6_minutes (data): Library: AAAAMMJJHHMN parsed ‘%Y%m%d%H%M’ (observation/api.py); observation/api.py converts the stamps of the overseas departments from FU to UTC with the offsets of _FU_HOURS_BY_DEPARTMENT and _FU_HOURS_BY_STATION (GH-2677), and takes the stamps of the other stations as UTC; request bounds: naive start/end -> UTC (model/request.py), filter is_between closed=’both’ (model/values.py); period files picked by the year of start/end moved to FU, the year of their stamps (observation/api.py). Source: “Les heures sont exprimées en UTC pour la métropole et en FU pour l’outre-mer” (French; dataset description on data.gouv.fr, identical for the 6-minute, hourly, daily and monthly datasets) = “Hours are expressed in UTC for metropolitan France and in FU [local time zone, “fuseau”] for overseas territories”. FU is not defined further in any file fetched. Field: “AAAAMMJJHHMN : date de la mesure (année mois jour heure minute)”, “RR : quantité de précipitation tombée en 6 minutes” (French) = “date of the measurement”, “precipitation fallen in 6 minutes” https://meteofrance.s3.sbg.io.cloud.ovh.net/data/synchro_ftp/BASE/MIN/MN_descriptif_champs.csv ; https://www.data.gouv.fr/datasets/donnees-climatologiques-de-base-6-minutes (dataset 6569ad61106d1679c93cdf77). Docs do not say start/end. Comparison: hourly RR1(H) == sum of 6-min stamps in (H-1h, H] for 299/299 wet hours (Paris 75114001, Apr-Oct 2025), vs 25/431 for [H, H+1h).

  • observation/hourly (core, others): Library: AAAAMMJJHH + ‘00’ parsed ‘%Y%m%d%H%M’ (observation/api.py); observation/api.py converts the stamps of the overseas departments from FU to UTC with the offsets of _FU_HOURS_BY_DEPARTMENT and _FU_HOURS_BY_STATION (GH-2677), and takes the stamps of the other stations as UTC; request bounds: naive start/end -> UTC (model/request.py), filter is_between closed=’both’ (model/values.py); period files picked by the year of start/end moved to FU, the year of their stamps (observation/api.py). Source: “Les heures sont exprimées en UTC pour la métropole et en FU pour l’outre-mer” (French; dataset description on data.gouv.fr, identical for the 6-minute, hourly, daily and monthly datasets) = “Hours are expressed in UTC for metropolitan France and in FU [local time zone, “fuseau”] for overseas territories”. FU is not defined further in any file fetched. Field: “AAAAMMJJHH : date de la mesure (année mois jour heure)”, “T : température sous abri instantanée”, “FF : force du vent moyenné sur 10 mn” (French) = “date of the measurement”, “instantaneous screen temperature”, “wind speed averaged over 10 min” https://meteofrance.s3.sbg.io.cloud.ovh.net/data/synchro_ftp/BASE/HOR/H_descriptif_champs.csv ; https://www.data.gouv.fr/datasets/donnees-climatologiques-de-base-horaires (dataset 6569b4473bedf2e7abad3b72). Docs say ‘instantanée’ only for T; FF and DD are ‘moyenné sur 10 mn’ (DD: ‘direction de FF’) and neither the field list nor the dataset description says which end of those 10 minutes the stamp is, so the wind is not filed as an instant and its position is unstated. Comparison: Paris 75114001 T/TD/U/PMER/PSTAT/FF/DD == SYNOP 07156 (UTC) at lag 0. Réunion 97418110 T == SYNOP 61980 only at hourly(t) = synop(t-4h), which is the FU offset converted. Overseas, FU is not defined in any file and the offsets are measured, not documented: T of every station that has a SYNOP station on the same site (postes_synop.geojson, UTC) equals SYNOP t at hourly(t) = synop(t - offset) with a mean absolute difference of 0.000 over 816 to 3,563 pairs per station in 2025 (each month with enough readings for 971-975, 984 and 985, so no DST there) and of at most 0.12 in 1997, 2005, 2008, 2012 and 2015 for 971-975, 984 and 985: 971 and 972 -4 h (Le Raizet, La Désirade, Saint-Barthélemy, Lamentin, Trinité-Caravelle), 973 -3 h (Cayenne, Saint-Laurent, Maripasoula, Saint-Georges), 974 +4 h (Gillot), 975 -4 h (Saint-Pierre, each month of 2025), 985 +3 h (Pamandzi), 986 +12 h (Hihifo), 987 -10 h (Faa’a, Bora-Bora, Hao, Hiva-Oa, Mangareva, Rapa, Takaroa, Taiarapu-Est), 988 +11 h (Nouméa, Koumac, La Tontouta, Surprise, Chesterfield); 984 differs by station: Glorieuses, Juan de Nova and Europa +3 h, Tromelin +4 h, Kerguelen and Nouvelle-Amsterdam +5 h, Dumont d’Urville +9 h, Crozet +5 h up to 2003-09-15 and +4 h from 2003-11-07 (the SYNOP archive has no reading of it in between; the library switches on 2003-10-01, so the stamps of those weeks may be an hour off). FU is thus not the legal time: 975 is -4 h (legally -3 h, -2 h in summer), Crozet +4 h (legally +5 h), Dumont d’Urville +9 h (legally +10 h), and the Marquesas and the Gambier are -10 h like Tahiti. SYNOP has 986-988 only since September 2025, so they are measured on September to December 2025; their mean diurnal cycle in the files of the 1990s to 2020-2024 stays within an hour of the 2025 one. The other stations of a department stay within one hour of its reference station’s mean diurnal cycle (2025-2026 file, every station with a full set of hours: 49, 46, 30, 55, 50 and 60 stations in 971, 972, 973, 974, 987 and 988), so the department’s offset is applied to all of them. Before 1996 nothing independent exists: the 3-hourly stamps of the 1960s to 1980s fall on the synoptic hours shifted by the offset (02, 05, 08 … h in 971, 975, 987 and 988), which fits but proves nothing, and Port Martin (98404006, 1950 to 1952) has no SYNOP station to compare with and is not converted (a warning says so). The 6-minute files are on the hourly clock: in every department the hourly RR1 equals the sum of the 6-minute RR stamped in the hour before it, in the file’s own stamps, for the stations compared (e.g. 97502001: 1,138 of 1,138 wet hours).

  • observation/hourly (core): Library: Same parse as the instant row (observation/api.py), no shift of the hour; observation/api.py converts the stamps of the overseas departments from FU to UTC with the offsets of _FU_HOURS_BY_DEPARTMENT and _FU_HOURS_BY_STATION (GH-2677), and takes the stamps of the other stations as UTC; request bounds: naive start/end -> UTC (model/request.py), filter is_between closed=’both’ (model/values.py); period files picked by the year of start/end moved to FU, the year of their stamps (observation/api.py). Source: “Les heures sont exprimées en UTC pour la métropole et en FU pour l’outre-mer” (French; dataset description on data.gouv.fr, identical for the 6-minute, hourly, daily and monthly datasets) = “Hours are expressed in UTC for metropolitan France and in FU [local time zone, “fuseau”] for overseas territories”. FU is not defined further in any file fetched. Fields: “RR1 : quantité de précipitation tombée en 1 heure”, “TN : température minimale sous abri dans l’heure”, “TX : température maximale sous abri dans l’heure”, “FXY : valeur maximale de FF dans l’heure” (French) = “precipitation fallen in 1 hour”, “min/max screen temperature within the hour”, “max of FF within the hour” (H_descriptif_champs.csv). Which hour is not stated. Comparison: RR1(H) == SYNOP 07156 rr1 (‘sur la dernière heure’) at lag 0 (mean|diff| 0.005) and == sum of 6-min stamps (H-1h, H]; SYNOP tn24/tx24/rr24 == hourly over (t-24h, t] 447/447. (https://www.data.gouv.fr/datasets/6569b4473bedf2e7abad3b72/)

  • observation/hourly (others): Library: Same parse (observation/api.py), no shift of the hour; observation/api.py converts the stamps of the overseas departments from FU to UTC with the offsets of _FU_HOURS_BY_DEPARTMENT and _FU_HOURS_BY_STATION (GH-2677), and takes the stamps of the other stations as UTC; request bounds: naive start/end -> UTC (model/request.py), filter is_between closed=’both’ (model/values.py); period files picked by the year of start/end moved to FU, the year of their stamps (observation/api.py). Source: “Les heures sont exprimées en UTC pour la métropole et en FU pour l’outre-mer” (French; dataset description on data.gouv.fr, identical for the 6-minute, hourly, daily and monthly datasets) = “Hours are expressed in UTC for metropolitan France and in FU [local time zone, “fuseau”] for overseas territories”. FU is not defined further in any file fetched. Fields: “GLO : rayonnement global horaire en heure UTC”, “INS : insolation horaire en heure UTC” (French) = “hourly global radiation in UTC hour”, “hourly sunshine in UTC hour” (H_descriptif_champs.csv and https://static.data.gouv.fr/resources/donnees-climatologiques-de-base-horaires/20260603-085646/climatologie-donnees-horaires-descriptif-20260603.pdf). Searched both files and the Confluence page linked from the dataset for start/end: silent; no second source compared. Overseas, GLO is on the FU clock, not in UTC as the source says: the mean GLO of the three stations with most readings per department peaks at the stamp hour 11 to 13 everywhere (Pierrefonds 97416463 at 12 where solar noon is 08:20 UTC, Saint-Pierre 97502001 at 12 where it is 15:40 UTC, Faa’a 98715002 at 12 where it is 22:00 UTC), so it is converted with the other columns.

  • observation/monthly (data): Library: AAAAMM parsed ‘%Y%m’ as day 1 00:00 UTC (observation/api.py); closed end bound means end=YYYY-MM-01 returns that whole month; observation/api.py parses the date column with a naive strptime and labels it 00:00 UTC for every department incl. 97x/98x; the FU date of an overseas station is kept, as a daily or monthly stamp names its day or month (GH-2677); request bounds: naive start/end -> UTC (model/request.py), filter is_between closed=’both’ (model/values.py); period buckets picked by calendar year of start/end (observation/api.py). Source: “Les heures sont exprimées en UTC pour la métropole et en FU pour l’outre-mer” (French; dataset description on data.gouv.fr, identical for the 6-minute, hourly, daily and monthly datasets) = “Hours are expressed in UTC for metropolitan France and in FU [local time zone, “fuseau”] for overseas territories”. FU is not defined further in any file fetched. “AAAAMM : mois”, “RR : cumul mensuel des hauteurs de précipitation”, “TX : moyenne mensuelle des températures maximales (TX) quotidiennes”, “TMM : moyenne mensuelle des températures moyennes (TM) quotidiennes” (French) = “month”, “monthly total precipitation”, “monthly mean of daily TX”, “monthly mean of daily TM” https://meteofrance.s3.sbg.io.cloud.ovh.net/data/synchro_ftp/BASE/MENS/MENSQ_descriptif_champs.csv ; https://www.data.gouv.fr/datasets/donnees-climatologiques-de-base-mensuelles (dataset 6569b3d7d193b4daf2b43edc). Comparison: monthly RR == sum of daily RR of the calendar month for Apr-Oct 2025 (7/7, Paris).

meteoswiss#

network

resolution

datasets

source time zone

reference time

verified

observation

10_minutes

all

UTC

end (of the preceding 10 minutes)

yes

observation

hourly

all

UTC

mixed (end of the hour since 2018; before 2018 the SYNOP schedule: stamp 16:00 covers 15:50 to 16:40)

yes

observation

daily

all

UTC

start (of the day; precipitation and snow 06:00 UTC to 06:00 UTC next day)

yes

observation

monthly

all

UTC

start (of the month)

yes

observation

annual

all

UTC

start (of the year)

yes

Notes:

  • observation/10_minutes: Library: reference_timestamp parsed ‘%d.%m.%Y %H:%M’ and labelled UTC unchanged (meteoswiss/observation/api.py); decade-chunked historical files (named by the decade of their UTC time stamps) chosen by comparing the decade with the UTC year of start/end (api.py); until GH-2680 the year of a bound was read as written, so a request such as 2010-01-01T00:30:00+02:00 (2009-12-31 22:30 UTC) chose the 2010-2019 file and returned no rows; output cut inclusively (model/values.py). Source: “All reference time stamps at MeteoSwiss are in UTC! Depending on the granularity the time stamp does define different intervals:” “t: The sum, mean or max/min of the last 10 minutes (ReferenceTS 16:00 = 15:50:01 to 16:00:00)” “for granularity t and h the time stamp defines the end of the measurement interval and for higher granularities (d, m and y) the time stamp defines the beginning of the interval!” https://opendatadocs.meteoswiss.ch/general/download

  • observation/hourly: Library: reference_timestamp parsed ‘%d.%m.%Y %H:%M’ and labelled UTC unchanged (meteoswiss/observation/api.py); decade-chunked historical files (named by the decade of their UTC time stamps) chosen by comparing the decade with the UTC year of start/end (api.py); until GH-2680 the year of a bound was read as written, so a request such as 2010-01-01T00:30:00+02:00 (2009-12-31 22:30 UTC) chose the 2010-2019 file and returned no rows; output cut inclusively (model/values.py). No distinction is made for pre-2018 hourly values. Source: “All reference time stamps at MeteoSwiss are in UTC! Depending on the granularity the time stamp does define different intervals:” “h: The sum, mean or max/min of the last six 10min-values (ReferenceTS 16:00 = 15:10 to 16:00). Please note: Hourly values before 2018 were calculated differently based on the SYNOP schedule (ReferenceTS 16:00 = 15:50 to 16:40)!” “for granularity t and h the time stamp defines the end of the measurement interval and for higher granularities (d, m and y) the time stamp defines the beginning of the interval!” https://opendatadocs.meteoswiss.ch/general/download

  • observation/daily: Library: reference_timestamp parsed ‘%d.%m.%Y %H:%M’ and labelled UTC unchanged (meteoswiss/observation/api.py); the files of this resolution have no decade in their names, so the request bounds choose no file by year (api.py); output cut inclusively (model/values.py). Source: “All reference time stamps at MeteoSwiss are in UTC! Depending on the granularity the time stamp does define different intervals:” “d: For most parameters the sum, mean or max/min from 00:00 to 23:50 of the according date. Exception for precipitation and snow (manual measurement times used for consistency) where the interval is 6:00 UTC until 5:50 UTC tomorrow (ReferenceTS 22.6.2023 = 22.6.2023 6:10 UTC to 23.6.2023 6:00 UTC)” “for granularity t and h the time stamp defines the end of the measurement interval and for higher granularities (d, m and y) the time stamp defines the beginning of the interval!” https://opendatadocs.meteoswiss.ch/general/download

  • observation/monthly: Library: reference_timestamp parsed ‘%d.%m.%Y %H:%M’ and labelled UTC unchanged (meteoswiss/observation/api.py); the files of this resolution have no decade in their names, so the request bounds choose no file by year (api.py); output cut inclusively (model/values.py). Source: “All reference time stamps at MeteoSwiss are in UTC! Depending on the granularity the time stamp does define different intervals:” “m: The sum, mean or max/min of the whole month from 1st to last day of month (ReferenceTS 1.6.2023 = 1.6.2023 00:10 UTC to 30.6.2023 24:00 UTC)” “for granularity t and h the time stamp defines the end of the measurement interval and for higher granularities (d, m and y) the time stamp defines the beginning of the interval!” https://opendatadocs.meteoswiss.ch/general/download

  • observation/annual: Library: reference_timestamp parsed ‘%d.%m.%Y %H:%M’ and labelled UTC unchanged (meteoswiss/observation/api.py); the files of this resolution have no decade in their names, so the request bounds choose no file by year (api.py); output cut inclusively (model/values.py). Source: “All reference time stamps at MeteoSwiss are in UTC! Depending on the granularity the time stamp does define different intervals:” “y: The sum, mean or max/min of the whole year (ReferenceTS 1.1.2023 = 1.1.2023 00:10 UTC to 31.12.2023 24:00 UTC)” “for granularity t and h the time stamp defines the end of the measurement interval and for higher granularities (d, m and y) the time stamp defines the beginning of the interval!” https://opendatadocs.meteoswiss.ch/general/download

metno#

network

resolution

datasets

source time zone

reference time

verified

frost

10_minutes

data

UTC

unverified (referenceTime = ‘observation time’; whether that is the end of the PT10M max/min/sum is not stated)

partly

frost

hourly

data

UTC

mixed (instant for air_temperature/dew point/humidity/pressure/cloud/snow; end for wind: the mean of the last 10 minutes before the stamp; PT1H mean radiation and sum precipitation stamped at ‘observation time’, start/end unstated)

partly

frost

6_hours

data

UTC

unverified (sum(precipitation_amount PT6H) stamped at ‘observation time’; start/end unstated)

partly

frost

daily

data

UTC

mixed (timeOffset PT0H: start, UTC calendar date; timeOffset != PT0H e.g. sum(precipitation_amount P1D) PT6H, max/min(air_temperature P1D) PT18H: end date, window ends at date+offset; snow_depth (surface_snow_thickness): the instant of the reading, stamped 00:00 UTC of its date)

yes

frost

monthly

data

UTC

mixed (timeOffset 0: start of calendar month; sum(precipitation_amount P1M) PT6H: last day of previous month 06 UTC to last day of month 06 UTC, labelled day 1)

yes

frost

annual

data

UTC

start (calendar year, referenceTime <year>-01-01T00:00; offset rules as monthly)

yes

Notes:

  • frost/10_minutes (data): Library: api.py parse referenceTime (’…Z’) as UTC unchanged; timeOffset is neither requested (no timeoffsets=default) nor read. Request bounds are converted to UTC and written with strftime(‘%Y-%m-%dT%H:%M:%S’) (api.py, _frost_time), which Frost reads as UTC; until GH-2681 they were written as the wall clock of the zone they were held in, so a bound in another zone named an instant its offset away; the end is sent a second later, as Frost’s end is exclusive (GH-2746), and the window then trimmed by the framework’s inclusive is_between. Source: “The Frost time specification is based on UTC time and ISO-8601.” (https://frost.met.no/concepts2.html); “Case 1: Observations with timeResolution shorter than 1 day … The referenceTime is then equal to the observation time, and no timeOffset should be added to referenceTime to get the original observation time.” (https://frost.met.no/concepts2.html). Data not fetched: no WD_AUTH__METNO_FROST credentials in this sandbox.

  • frost/hourly (data): Library: api.py parse referenceTime (’…Z’) as UTC unchanged; timeOffset is neither requested (no timeoffsets=default) nor read. Request bounds are converted to UTC and written with strftime(‘%Y-%m-%dT%H:%M:%S’) (api.py, _frost_time), which Frost reads as UTC; until GH-2681 they were written as the wall clock of the zone they were held in, so a bound in another zone named an instant its offset away; the end is sent a second later, as Frost’s end is exclusive (GH-2746), and the window then trimmed by the framework’s inclusive is_between. Source: “The Frost time specification is based on UTC time and ISO-8601.” (https://frost.met.no/concepts2.html); “Case 1: Observations with timeResolution shorter than 1 day … The referenceTime is then equal to the observation time, and no timeOffset should be added to referenceTime to get the original observation time.” (https://frost.met.no/concepts2.html). Wind: hourly wind_speed and wind_from_direction are named wind_speed_mean_last_10min and wind_direction_mean_last_10min and are Frost’s mean over the 10 minutes before the observation time, so the stamp is the end of their window, not an instant (Frost’s element descriptions, as in docs/data/provider/metno/frost/hourly.md: “Mean wind speed is registered as a mean value of the wind speed over the last ten minutes before the observation time.”; “Mean wind direction over the last ten minutes before the observation time.”). Data not fetched: no credentials.

  • frost/6_hours (data): Library: api.py parse referenceTime (’…Z’) as UTC unchanged; timeOffset is neither requested (no timeoffsets=default) nor read. Request bounds are converted to UTC and written with strftime(‘%Y-%m-%dT%H:%M:%S’) (api.py, _frost_time), which Frost reads as UTC; until GH-2681 they were written as the wall clock of the zone they were held in, so a bound in another zone named an instant its offset away; the end is sent a second later, as Frost’s end is exclusive (GH-2746), and the window then trimmed by the framework’s inclusive is_between. Source: “The Frost time specification is based on UTC time and ISO-8601.” (https://frost.met.no/concepts2.html); “Case 1: Observations with timeResolution shorter than 1 day … The referenceTime is then equal to the observation time, and no timeOffset should be added to referenceTime to get the original observation time.” (https://frost.met.no/concepts2.html). Data not fetched: no credentials.

  • frost/daily (data): Library: api.py parse referenceTime (’…Z’) as UTC, a daily stamp then truncated to the day (see below); timeOffset is neither requested (no timeoffsets=default) nor read. Request bounds are converted to UTC and written with strftime(‘%Y-%m-%dT%H:%M:%S’) (api.py, _frost_time), which Frost reads as UTC; until GH-2681 they were written as the wall clock of the zone they were held in, so a bound in another zone named an instant its offset away; the window then trimmed by the framework’s inclusive is_between. A station with several timeOffsets for one element yields several rows with the same date stamp. Frost stores the instantaneous snow depth (surface_snow_thickness) with the time of its reading as its referenceTime; api.py truncates a daily stamp to the day, so it is stamped 00:00 UTC of the date of its reading like the other daily parameters (GH-2729). The reading of the observation time on a date belongs to that date. A request to Frost for daily values ends at 00:00 UTC of the date after the date of its end, as “Intervals in Frost are open-ended: [t0,t1>, i.e. representing times later than or equal to t0 and earlier than t1” (https://frost.met.no/concepts2.html), so a request whose end is 00:00 UTC of a date asks for the whole of that date, the reading of 06:00 UTC included (GH-2729, GH-2746); the framework trims what lies beyond the window. This follows the documentation and is not verified against Frost: no key was available. Source: “The Frost time specification is based on UTC time and ISO-8601.” (https://frost.met.no/concepts2.html); “Diurnal data … are stored with date only in the data storage with a referenceTime of the form <year>-<month>-<day>T00:00. If timeOffset = PT0H, the observation value represents the period from midnight to midnight on the calendar date … If timeOffset != PT0H, the observation represents the last 24 hr, at the time after midnight equal to timeOffset. E.g. elementId = sum(precipitation_amount P1D) has timeOffset PT6H. This means that the observation period lasts continuously from the previous date at 06 UTC to current date 06 UTC.” “Instantaneous observations with timeResolution ≥ 1 day are stored with date and time … The referenceTime is equal to the observation time” (https://frost.met.no/concepts2.html); ‘max(air_temperature P1D) P1D PT18H Standard sampling period for temperature extremes is 19-18 UTC’, ‘sum(precipitation_amount P1D) P1D PT6H Standard sampling period for precipitation is 07-06 UTC’, ‘Sometimes multiple timeseries are available at the same station, with the only difference being the timeOffset value.’ (https://frost.met.no/dataclarifications.html). Data not fetched: no credentials.

  • frost/monthly (data): Library: api.py parse referenceTime (’…Z’) as UTC unchanged; timeOffset is neither requested (no timeoffsets=default) nor read. Request bounds are converted to UTC and written with strftime(‘%Y-%m-%dT%H:%M:%S’) (api.py, _frost_time), which Frost reads as UTC; until GH-2681 they were written as the wall clock of the zone they were held in, so a bound in another zone named an instant its offset away; the end is sent a second later, as Frost’s end is exclusive (GH-2746), and the window then trimmed by the framework’s inclusive is_between. Source: “The Frost time specification is based on UTC time and ISO-8601.” (https://frost.met.no/concepts2.html); “If timeOffset = 0, the monthly value represents the calendar month, midnight to midnight. The referenceTime is of the form <year><month>-01T00:00, i.e. the start of the observation period. If timeOffset != 0, the monthly value represents the period from p1 to p2 where p1 and p2 are midnight + timeOffset on the last day of the previous and current month respectively.” (https://frost.met.no/concepts2.html). Data not fetched: no credentials.

  • frost/annual (data): Library: api.py parse referenceTime (’…Z’) as UTC unchanged; timeOffset is neither requested (no timeoffsets=default) nor read. Request bounds are converted to UTC and written with strftime(‘%Y-%m-%dT%H:%M:%S’) (api.py, _frost_time), which Frost reads as UTC; until GH-2681 they were written as the wall clock of the zone they were held in, so a bound in another zone named an instant its offset away; the end is sent a second later, as Frost’s end is exclusive (GH-2746), and the window then trimmed by the framework’s inclusive is_between. Source: “The Frost time specification is based on UTC time and ISO-8601.” (https://frost.met.no/concepts2.html); “Annual data cover the calendar year. ReferenceTime and timeOffset for annual data follow the same conventions as for seasonal and monthly data.” (https://frost.met.no/concepts2.html). Data not fetched: no credentials.

metoffice#

network

resolution

datasets

source time zone

reference time

verified

observation

daily

rain

UTC (GMT)

end (the 24 h, usually 09-09Z, ending on the stated day)

yes

observation

daily

temperature

UTC (GMT)

mixed (periods 09-09Z, 09-21Z, 21-09Z, each identified by ob_end_time and ob_hour_count)

yes

observation

daily

weather

UTC (GMT)

mixed (sunshine: the day 0000-2359Z; snow depth and fresh snow: spot values usually at 0900Z)

yes

observation

hourly

rain

UTC (GMT)

end (of the hour, or hours, ending at ob_end_time)

yes

observation

hourly

weather

UTC (GMT)

instant (ob_time)

yes

observation

hourly

wind

UTC (GMT)

end (mean over ob_hour_count hours ending at ob_end_time; hourly and 24 h 09-09Z means in one table)

yes

observation

hourly

radiation

UTC (GMT)

end (hour ending at ob_end_time; 24 h values usually 0000-2359Z in the same table)

yes

observation

hourly

soil_temperature

UTC (GMT)

instant (ob_time; daily readings usually at 0900Z, hourly ones usually at 10 cm)

yes

Notes:

  • observation/daily (rain): Library: `ob_date` parsed as UTC and truncated to the day; rows with `ob_day_cnt` != 1 dropped; max per day (src/wetterdienst/provider/metoffice/observation/parser.py; api.py). Data not fetched: CEDA token not configured. Source: “This entity contains rainfall amounts measured over one or more 24 hour periods, ending on the stated day.” and “Daily rainfall over a 24 hour period - usually 09-09Z.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/RD_Table.html); “The standard of time for all UK observations is GMT. […] The meteorological convention for midnight is 0000 hours on the new day.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/ukmo_guide.html, section 2.4)

  • observation/daily (temperature): Library: `ob_end_time` parsed as UTC, truncated to the calendar day, then max (min for min_air_temp/min_grss_temp) over all rows ending that day, regardless of `ob_hour_count` (src/wetterdienst/provider/metoffice/observation/parser.py; api.py). A day D thus combines the 09-21Z max of D with the 09-09Z value ending 09Z on D (D-1’s day). Read from the files (25 stations, 2022): the daily temperature file mixes 12 h rows ending 09:00 (night) and 21:00 (day) with 24 h rows ending 09:00, 2,962 of its 14,334 rows being the latter, so a 24 h 09-09Z row is stamped the day after most of its coverage. Kept as it is: the stamp is the date of `ob_end_time`, the source’s 09Z climatological day. Source: “ob_end_time and ob_hour_count define the observation period.” and “For air temperatures, the observation period will usually be one of the following: 09-09Z, 09-21Z, 21-09Z.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/TD_Table.html); “values of rainfall, maximum temperature, run of wind and sunshine are “thrown back” to the previous day […] This practice does not affect the way the data are stored in MIDAS as the database indicates the observation period by the two parameters: observation end time and hour count.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/ukmo_guide.html); “The standard of time for all UK observations is GMT. […] The meteorological convention for midnight is 0000 hours on the new day.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/ukmo_guide.html, section 2.4)

  • observation/daily (weather): Library: `ob_end_time` parsed as UTC and truncated to the day, max per day (src/wetterdienst/provider/metoffice/observation/parser.py; api.py): a 09Z spot value is stamped 00:00 of its day. Source: “Sunshine duration, snow day, hail day, thunder day and gale day are usually for the period 0000-2359Z, i.e. ob_end_time = ‘2359’ and ob_hour_count = 24 . Concrete state, lying snow, snow depth, fresh snow depth and fresh mountain snowfall are “spot” values, usually at 0900Z, i.e. ob_end_time = 0900 and ob_hour_count = 0 .” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/WD_Table.html); “The standard of time for all UK observations is GMT. […] The meteorological convention for midnight is 0000 hours on the new day.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/ukmo_guide.html, section 2.4)

  • observation/hourly (rain): Library: `ob_end_time` parsed as UTC, truncated to the hour, max per hour; only rows with `ob_hour_count` 1 are kept, as the 12 h rows (09-21Z, 21-09Z) ending 09:00/21:00 (719 of 9,299 rows of Tyndrum 2023) competed with the 1 h value of that hour and won (src/wetterdienst/provider/metoffice/observation/parser.py; api.py; GH-3137). Source: “This entity contains rainfall amount (and duration from tilting syphon gauges) during the hour (or hours) ending at the specified time.” and “Hourly rainfall ( ob_hour_count = 1) 09-21Z or 21-09Z rainfall from NCM and F7113 ( ob_hour_count = 12)” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/RH_Table.html); “The standard of time for all UK observations is GMT. […] The meteorological convention for midnight is 0000 hours on the new day.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/ukmo_guide.html, section 2.4)

  • observation/hourly (weather): Library: `ob_time` parsed as UTC, truncated to the hour, max per hour (src/wetterdienst/provider/metoffice/observation/parser.py; api.py). Source: “ob_time Date and time of observation” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/WH_Table.html); “The standard of time for all UK observations is GMT. […] The meteorological convention for midnight is 0000 hours on the new day.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/ukmo_guide.html, section 2.4)

  • observation/hourly (wind): Library: `ob_end_time` parsed as UTC, truncated to the hour, max per hour; only rows with `ob_hour_count` 1 are kept (all 209,010 rows of 25 station-years of 2022 are, so nothing changes) (src/wetterdienst/provider/metoffice/observation/parser.py; api.py). Source: “ob_end_time Data and time at end of observation” and “Mean value of wind speed for a 24 hour period, usually 09-09Z, obtained from run-of-wind.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/WM_Table.html); “The standard of time for all UK observations is GMT. […] The meteorological convention for midnight is 0000 hours on the new day.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/ukmo_guide.html, section 2.4)

  • observation/hourly (radiation): Library: `ob_end_time` parsed as UTC, truncated to the hour, max per hour; only rows with `ob_hour_count` 1 are kept, as the 24 h totals (364 of 9,084 rows of Kinloss 2023) were stamped at 23:00 and won (src/wetterdienst/provider/metoffice/observation/parser.py; api.py; GH-3137). Source: “In all cases, ob_end_time and ob_hour_count define the observation period. Values are either for an hour or for 24 hours.” and “Daily - 24 hour values are usually for 0000-2359Z on the stated day.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/RO_Table.html); “The standard of time for all UK observations is GMT. […] The meteorological convention for midnight is 0000 hours on the new day.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/ukmo_guide.html, section 2.4)

  • observation/hourly (soil_temperature): Library: `ob_time` parsed as UTC, truncated to the hour, max per hour (src/wetterdienst/provider/metoffice/observation/parser.py; api.py). Source: “Daily soil temperatures are usually measured at 0900Z.” and “Hourly soil temperatures are usually at 10 cm.” and “ob_time Date and time of observation” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/ST_Table.html); “The standard of time for all UK observations is GMT. […] The meteorological convention for midnight is 0000 hours on the new day.” (https://artefacts.ceda.ac.uk/badc_datadocs/ukmo-midas/ukmo_guide.html, section 2.4)

smhi#

network

resolution

datasets

source time zone

reference time

verified

observation

1_minute

data

UTC

mixed (instant for temperature/humidity/pressure/snow depth; 1-minute mean for wind/visibility; precipitation ‘under en minut’, stamp position unstated)

partly

observation

hourly

data

UTC

mixed (instant for temperature/dew point/humidity/pressure/cloud/visibility; wind (wind_speed_mean_last_10min, wind_direction_mean_last_10min): a 10-min mean, its position relative to the stamp not stated; hourly gust max and hourly precipitation sum: stamp position unstated (observed: consistent with end))

partly

observation

daily

data

UTC

start (temperature_air_mean_2m: UTC day 00:00:01-24:00 labelled with that day)

yes

observation

daily

data

UTC

start (precipitation_amount: sum 06:00 UTC D to 06:00 UTC D+1, labelled D)

yes

observation

daily

data

UTC

end (temperature_air_max_2m/min_2m: 18:00 UTC D-1 to 18:00 UTC D, labelled D)

yes

observation

daily

data

UTC

instant (snow_depth: the reading at 06:00 UTC of D, stamped D 00:00 UTC)

yes

observation

monthly

data

UTC

start (temperature_air_mean_2m: UTC calendar month; precipitation_amount: 06:00 UTC on the 1st to 06:00 UTC on the 1st of next month)

yes

Notes:

  • observation/1_minute (data): Library: parser.py builds the stamp from ‘Datum’+’Tid (UTC)’ (sub-daily) or ‘Representativt dygn’ / ‘Representativ månad’+’-01’ (daily/monthly) and tags it UTC with replace_time_zone; no shift. Both periods (corrected-archive, latest-months) are always fetched and de-duplicated on timestamp (api.py); request bounds only via the framework’s inclusive is_between (model/values.py). Only the rolling ‘latest-day’ period is fetched for 1_minute. Source: Data-file header ‘Datum;Tid (UTC);Lufttemperatur;Kvalitet’ with description ‘Lufttemperatur;momentanvärde varje minut’ (Swedish: ‘instantaneous value every minute’); parameter list summaries ‘1 minuts medel varje minut’ (wind 47/48, visibility 51) and ‘Nederbörd under en minut’ (46) (https://opendata-download-metobs.smhi.se/api/version/1.0/parameter/45/station/98230/period/latest-day/data.csv, https://opendata-download-metobs.smhi.se/api/version/1.0.json). API docs only say: ‘All dates in the JSON answers are in Unix time stamp.’ (https://opendata.smhi.se/metobs/resources/data)

  • observation/hourly (data): Library: parser.py builds the stamp from ‘Datum’+’Tid (UTC)’ (sub-daily) or ‘Representativt dygn’ / ‘Representativ månad’+’-01’ (daily/monthly) and tags it UTC with replace_time_zone; no shift. Both periods (corrected-archive, latest-months) are always fetched and de-duplicated on timestamp (api.py); request bounds only via the framework’s inclusive is_between (model/values.py). Source: Header ‘Datum;Tid (UTC);Lufttemperatur;Kvalitet’ with ‘Lufttemperatur;momentanvärde, 1 gång/tim’ (instantaneous, once per hour) and ‘Nederbördsmängd;summa 1 timme, 1 gång/tim’ (sum 1 hour) (https://opendata-download-metobs.smhi.se/api/version/1.0/parameter/1/station/98230/period/latest-months/data.csv, https://opendata-download-metobs.smhi.se/api/version/1.0/parameter/7/station/98230/period/latest-months/data.csv). Wind: parameters 3 and 4 read ‘Vindriktning’ and ‘Vindhastighet’ with ‘medelvärde 10 min, 1 gång/tim’ (mean of 10 min, once per hour; https://opendata-download-metobs.smhi.se/api/version/1.0/parameter/3.json and /4.json), which does not say which end of the 10 minutes the stamp is, so it is not filed as an instant. Docs do not say whether a sum/mean is stamped at its start or end; the JSON period of latest-day begins at xx:00:01 and its first value is stamped one hour later (observed 2026-10-10), which fits end-stamping but is not a statement.

  • observation/daily (data): Library: parser.py builds the stamp from ‘Datum’+’Tid (UTC)’ (sub-daily) or ‘Representativt dygn’ / ‘Representativ månad’+’-01’ (daily/monthly) and tags it UTC with replace_time_zone; no shift. Both periods (corrected-archive, latest-months) are always fetched and de-duplicated on timestamp (api.py); request bounds only via the framework’s inclusive is_between (model/values.py). Stamped at 00:00 UTC of ‘Representativt dygn’. Source: ‘Lufttemperatur;medelvärde 1 dygn, 1 gång/dygn, kl 00’ and rows ‘Från Datum Tid (UTC);Till Datum Tid (UTC);Representativt dygn’ -> ‘2026-06-02 00:00:01;2026-06-03 00:00:00;2026-06-02’ (https://opendata-download-metobs.smhi.se/api/version/1.0/parameter/2/station/98230/period/latest-months/data.csv)

  • observation/daily (data): Library: parser.py builds the stamp from ‘Datum’+’Tid (UTC)’ (sub-daily) or ‘Representativt dygn’ / ‘Representativ månad’+’-01’ (daily/monthly) and tags it UTC with replace_time_zone; no shift. Both periods (corrected-archive, latest-months) are always fetched and de-duplicated on timestamp (api.py); request bounds only via the framework’s inclusive is_between (model/values.py). Stamped at 00:00 UTC of ‘Representativt dygn’ (6 h before the period starts). Source: ‘Nederbördsmängd;summa 1 dygn, 1 gång/dygn, kl 06’ with row ‘2026-06-01 06:00:01;2026-06-02 06:00:00;2026-06-01’ (Från/Till Datum Tid (UTC); Representativt dygn) (https://opendata-download-metobs.smhi.se/api/version/1.0/parameter/5/station/98230/period/latest-months/data.csv)

  • observation/daily (data): Library: parser.py builds the stamp from ‘Datum’+’Tid (UTC)’ (sub-daily) or ‘Representativt dygn’ / ‘Representativ månad’+’-01’ (daily/monthly) and tags it UTC with replace_time_zone; no shift. Both periods (corrected-archive, latest-months) are always fetched and de-duplicated on timestamp (api.py); request bounds only via the framework’s inclusive is_between (model/values.py). Stamped at 00:00 UTC of ‘Representativt dygn’ = the day the 24 h window ends. Source: ‘Lufttemperatur;min, 1 gång per dygn’ with row ‘2026-06-01 18:00:01;2026-06-02 18:00:00;2026-06-02’ (https://opendata-download-metobs.smhi.se/api/version/1.0/parameter/19/station/98230/period/latest-months/data.csv); same layout for max (https://opendata-download-metobs.smhi.se/api/version/1.0/parameter/20/station/98230/period/latest-months/data.csv)

  • observation/daily (data): Library: parser.py builds the stamp from ‘Datum’+’Tid (UTC)’ (sub-daily) or ‘Representativt dygn’ / ‘Representativ månad’+’-01’ (daily/monthly) and tags it UTC with replace_time_zone; no shift. Both periods (corrected-archive, latest-months) are always fetched and de-duplicated on timestamp (api.py); request bounds only via the framework’s inclusive is_between (model/values.py). This file has ‘Datum’+’Tid (UTC)’ and the parser gives the instant of the reading, 06:00 UTC; api.py truncates a daily stamp to the day, so snow_depth is stamped 00:00 UTC of the date of its reading like the other daily parameters, and a one-day request (start=end=date) returns it (GH-2683, it was dropped while the stamp was 06:00 UTC). The reading of 06:00 UTC on a date belongs to that date. Source: ‘Snödjup;momentanvärde, 1 gång/dygn, kl 06’ with header ‘Datum;Tid (UTC);Snödjup’ and row ‘2026-06-02;06:00:00;0.00’ (https://opendata-download-metobs.smhi.se/api/version/1.0/parameter/8/station/98230/period/latest-months/data.csv)

  • observation/monthly (data): Library: parser.py builds the stamp from ‘Datum’+’Tid (UTC)’ (sub-daily) or ‘Representativt dygn’ / ‘Representativ månad’+’-01’ (daily/monthly) and tags it UTC with replace_time_zone; no shift. Both periods (corrected-archive, latest-months) are always fetched and de-duplicated on timestamp (api.py); request bounds only via the framework’s inclusive is_between (model/values.py). Stamped at 00:00 UTC on day 1 of ‘Representativ månad’. Source: ‘Lufttemperatur;medel, 1 gång per månad’ row ‘2026-06-01 00:00:01;2026-07-01 00:00:00;2026-06’ (https://opendata-download-metobs.smhi.se/api/version/1.0/parameter/22/station/98230/period/latest-months/data.csv); ‘Nederbördsmängd;summa, 1 gång per månad’ row ‘2026-06-01 06:00:01;2026-07-01 06:00:00;2026-06’ (https://opendata-download-metobs.smhi.se/api/version/1.0/parameter/23/station/98230/period/latest-months/data.csv)