Phenology

Phenology#

Overview#

The DWD phenological network records the calendar of plant development: for one plant at one station in one reference year, the day (DWD’s Jultag, see Reading Jultag) on which a developmental phase was reached — hazel flowering, beech leaf unfolding, winter wheat harvest. Around 1200 volunteer observers report from roughly 6600 stations, and the series reach back to 1925 for many of the wild plants.

Observations come from two reporter groups, which are kept apart by the dataset name prefix:

  • annual reporters (Jahresmelder, annual_*) — the full observation programme, submitted once the season is over. ~6600 stations, 72 plant datasets.

  • immediate reporters (Sofortmelder, immediate_*) — a smaller set of phases, reported within days of the observation. ~1200 stations, 38 plant datasets.

Data is published at observations_germany/phenology, one semicolon-separated text file per plant, reporter group and period, with the station catalogues in CDC/help. No authentication is required. See the DWD’s phenology pages for the observation guidelines behind the phase definitions.

Data model#

Phenology carries one dimension the rest of the library does not — the plant. It maps onto the file layout:

  • resolution — annual, one observation per phase and reference year.

  • dataset — one plant, as observed by one reporter group, which is exactly one source file: annual_common_hazel, immediate_winter_wheat, annual_grapevine_late_ripening.

  • parameter — one phenological phase: phenology_flowering_beginning, phenology_leaf_unfolding_beginning, phenology_harvest. The original name is DWD’s numeric Phase_id, which is the only name the data files themselves use.

  • value — Jultag, which DWD describes as the day of the year the phase was reached, dimensionless. It is returned as DWD writes it, and it is the day of the year only for a date in the reference year, and in a leap year since about 1952 only before 29 February; see Reading Jultag below.

  • timestamp — the 1st of January of the reference year, not the date of the phase. The entry date (Eintrittsdatum) that DWD also publishes is not carried; it can mostly, but not always, be worked out from the stamp and Jultag.

  • quality_flags — Qualitaetsniveau, as DWD writes it: 1 formal check only, 7 checked in routine but uncorrected, 10 checked and corrected.

from wetterdienst.provider.dwd.phenology import DwdPhenologyRequest

request = DwdPhenologyRequest(
    parameters=[("annual", "annual_common_hazel", "phenology_flowering_beginning")],
    periods="recent",
)
values = next(request.filter_by_station_id("07521").values().query()).df

Reading Jultag#

DWD’s description files call Jultag the “day of the year” (DDD) and say nothing more about how it is counted. The data show that it is the day of the year only for a date in the reference year, and in a leap year since about 1952 only before 29 February; the library returns it as DWD writes it. Checked on the current and historical files of winter wheat and winter rape (annual reporters), hazel (both reporter groups) and winter wheat and winter rape (immediate reporters), 2026-10-13:

  • Leap years. From about 1952 the count is of a 365-day year: 29 February is 59, as 28 February is, and 1 March is 60, so from 29 February Jultag is one below the real day of the year. Winter wheat, historical: 175,633 of 176,470 March-to-December rows of the leap years since 1952 (the rest, mostly 1952 to 1968, carry the real day of the year). Hazel, historical and recent: every March-to-December row of a leap year (1,243 and 169). Before 1952 the leap years carry the real day of the year (154 of 159 winter wheat rows).

  • A date before the reference year. A phase can fall before 1 January of the Referenzjahr: DWD’s observation handbook names the flowering of hazel, alder and snowdrop, and the beginning of growth in height of winter rape (phase 67). It is written as a negative number of days before 1 January: 31 December is -1 and there is no 0 (20231229 of Referenzjahr 2024 is -3; 243 of 243 such rows of the immediate-reporter hazel historical file, 63 of 63 of the recent one, 508 of 508 and 70 of 70 of the annual-reporter hazel). In the annual-reporter crop files the exception is the earlier year’s positive day of the year, which looks like a day of the reference year: all 37 rows of Referenzjahr 1963 dated in 1962 in winter wheat (19621010 is 283; winter rape has 20) and two of 2024 in winter wheat (20231203 is 337). A date in the year after the reference year runs on past 365 (20170115 of 2016 is 380; three winter wheat rows).

  • Working out the entry date. The timestamp is the 1st of January of the Referenzjahr and the entry date is not carried, so add Jultag - 1 days to the timestamp for a Jultag up to 59 (or in a year that is not a leap year), Jultag days for a Jultag of 60 or more in a leap year, and a negative Jultag as it is (-3 gives 29 December of the year before). That reproduces the entry date of nearly every row of the files above, apart from three kinds: 28 and 29 February, which share 59; a row of an older leap year that carries the real day of the year, which comes out one day late (a Jultag of 100 in 1960 may be 9 or 10 April); and the positive earlier-year days above, which come out in the wrong year. A few dozen rows (Referenzjahr 2013, station 12656, 20130510 is 257) follow none of it.

Periods#

  • historical — the full archive, re-released about once a year. DWD leaves the previous releases in place beside the current one, distinguished only by the end year in the file name; only the latest is complete, and that is the one read.

  • recent — the current year and the three before it.

Not every plant has both. Fifteen datasets are historical only — those discontinued before the recent files begin, among them spring wheat, tomato, white cabbage, the two farming activities, three of the five annual-reporter grapevine varieties and all three immediate-reporter ones. Two are recent only, annual_beet and annual_currant_all_varieties, having been added since the last historical release; their whole record lives in the recent file, so a date range reaching back before the recent window still reads them.

Warning

The files are per plant, not per station: asking for one station’s historical series downloads the file holding every station, which for the larger crop and tree series is 50–160 MB. A request parses each file once and keeps the stations it asked for, so a many-station request costs no more than a one-station request — but a one-station request is not cheap.

License#

The German Weather Service specified their data as being open though they ask you to reference them as copyright owner. Take a look at the Open Data Strategy at the DWD and the Official Copyright statements before using the data.