Install as pydmi-weather, import as pydmi.
Python wrapper for the free Open Data APIs from DMI (Danmarks Meteorologiske Institut).
| API | Status |
|---|---|
| Forecast Data EDR API | ✅ DMIForecast / DMIForecastAsync |
| Meteorological Observations API | ✅ DMIObservations / DMIObservationsAsync |
| Lightning Data API | ✅ DMILightning / DMILightningAsync |
- HARMONIE DINI model: ~2 km grid, hourly steps, ~60 h ahead, new run every 3 hours
- Sync client: standard library only, thread-safe. Use it in services like WeatherDataLogger.
- Async client: aiohttp, accepts an injected session. Use it in Home Assistant.
- Output in normal units (°C, hPa, m/s, mm, %) with unit-suffixed field names
- Derived Home Assistant condition strings, day/night aware
- Daily aggregates in your local time zone
No API key is needed. DMI removed authentication from opendataapi.dmi.dk.
pip install pydmi-weather # sync client, no dependencies
pip install "pydmi-weather[async]" # + aiohttp for the async clientfrom pydmi import DMIForecast, ATTRIBUTION
dmi = DMIForecast(55.64, 12.08) # lat, lon
fc = dmi.get_forecast() # cached for update_interval (30 min)
print(fc.model_run, ATTRIBUTION)
now = fc.current()
print(now.temperature_c, now.condition, now.wind_speed_ms)
for h in fc.hourly()[:6]:
print(h.datetime, h.temperature_c, h.precipitation_mm, h.condition)
for d in fc.daily(): # Europe/Copenhagen by default
print(d.date, d.temperature_min_c, d.temperature_max_c, d.precipitation_mm, d.condition)
payload = fc.to_dict() # JSON-ready, e.g. for MQTTAsync:
from pydmi.client_async import DMIForecastAsync
async with DMIForecastAsync(55.64, 12.08, session=session) as dmi:
fc = await dmi.get_forecast()See examples/ for an MQTT publisher (WeatherDataLogger style) and a Home Assistant
coordinator and weather entity.
from pydmi import DMIObservations
obs = DMIObservations(55.74, 12.15).get_observations() # cached for 10 min
print(obs.temperature_c, obs.wind_speed_ms, obs.global_radiation_wm2)
print(obs.values["temperature_c"].station_name, obs.values["temperature_c"].observed)
payload = obs.to_dict() # values + where/when each came fromFor each measurement, the nearest active Synop station that measures it is used (within 50 km). That can be different stations for different values: near Roskilde, temperature and wind come from Roskilde Lufthavn, solar radiation from Sjælsmark. The station list is fetched once a day, then one request per station every 10 minutes, 2 seconds apart. A value older than one hour counts as missing. If one station fails, values from the others are still returned.
Fields: temperature_c, dew_point_c, humidity_pct, pressure_hpa (sea level), wind_speed_ms
(10-min mean), wind_bearing_deg, wind_gust_ms (highest gust last 10 min), precipitation_10min_mm,
precipitation_1h_mm, cloud_cover_pct, visibility_m, global_radiation_wm2, sunshine_10min_min.
The async client is pydmi.client_async.DMIObservationsAsync.
from pydmi import DMILightning
lt = DMILightning(55.74, 12.15, radius_km=50).get_lightning() # cached for 5 min
print(lt.count_1h, lt.count_10min)
if lt.nearest:
print(lt.nearest.distance_km, "km at", lt.nearest.bearing_deg, "deg")
if lt.latest:
print(lt.latest.observed, lt.latest.current_ka, "kA")One request every 5 minutes for the last hour of strikes in a box around the location; strikes
outside the radius (default 50 km, max 300 km) are dropped. bearing_deg is the direction from your
location to the strike (0 = north, 90 = east). type is DMI's lightning type code, passed through
unchanged (DMI does not document its values); all types are counted.
DMI Open Data is free to use, and the library is built to keep it that way:
| What | How |
|---|---|
| Fair use limit (500 req / 5 s, shared by everyone) | Responses are cached. update_interval defaults to 30 min and cannot go below 10 min. The model only changes every 3 h anyway. |
| HTTP 429 | The client honours Retry-After. While it waits it returns the cached forecast and does not contact DMI. |
| Minimal payload | One /position request, 13 parameters, CoverageJSON (DMI says it is >10× smaller than GeoJSON), gzip |
| Cacheable requests | Coordinates are rounded to 2 decimals (~1 km, finer than the 2 km grid) |
| Identification | User-Agent: pydmi/<version> |
| Concurrency | A lock makes concurrent callers share one request |
DMI's data is licensed under CC BY 4.0.
Wherever the data is shown, you must credit DMI. Use pydmi.ATTRIBUTION
(fc.attribution). In Home Assistant, set it as _attr_attribution.
- Precipitation: DMI's
total-precipitationis accumulated from the start of the run. The library converts it to the amount per hour.precipitation_mmon an hour is the amount expected during that hour (the Home Assistant / met.no convention). The last step isNone. - Condition: the EDR API has no weather symbol, so one is derived. Priority order:
thunder (lightning probability ≥ 30 %; note DMI returns this in percent, although its metadata says 0..1), precipitation (≥ 0.3 mm/h, with rain, snow,
sleet or hail taken from
precipitation-type), fog (visibility < 1 km), wind (≥ 13.9 m/s), then cloud cover. Night usesclear-night. All thresholds are inconst.py. - Daily: grouped by local date. Today covers the rest of the day. Later days with fewer than 12 hours (the end of the run) are dropped.
- Model run:
fc.model_runis the first time step of the run (UTC). - Chance of precipitation:
precipitation_probability_pct(0–100, in steps of 10) is the chance of at least 0.1 mm during the hour. It comes from DMI's ensemble (harmonie_dini_eps_percentiles): the spread of 1-hour precipitation over the ensemble members is read as a distribution, and the chance of exceeding 0.1 mm is interpolated from it. The daily value is the highest hourly chance that day (an approximation). The ensemble is fetched at most once an hour (it changes every few hours) and reused for every forecast update in between, 2 seconds after the forecast request. If it fails, the forecast is still returned withNonehere, and only the ensemble is retried after 5 minutes (client.probability_pendingisTruemeanwhile). The ensemble request may take up to 60 seconds, and a warning is only logged after 3 failed attempts in a row (before that, at debug level). Turn it off withinclude_precipitation_probability=False. - Current conditions:
fc.current()interpolates instantaneous values (temperature, pressure, humidity, wind, cloud cover, radiation) linearly between the two hourly steps around now, so values change smoothly through the hour. Wind direction takes the shortest way around the compass. Precipitation is the amount for the current hour. Usefc.current(interpolate=False)to get the start-of-hour step unchanged.
pip install -e ".[test]"
pytestThe tests use a local mock server, so DMI is never contacted.