Provider capabilities
Use a local instrument for the readings it actually measures. Use other stations and weather services to fill gaps, with their age and location in view.
A provider’s capability does not guarantee that a reading is present now. Hardware options, coverage, station reporting, credentials, and failures all affect availability.
Understand the source label
| Type | What the value represents |
|---|---|
| Measured | An instrument observation at a station. A nearby station may differ from your yard. |
| Radar | An estimate derived from radar observations, sometimes adjusted with gauges. |
| Nowcast | A short-range analysis or estimate of current conditions. |
| Model | Computed current conditions rather than a direct instrument reading. |
| Forecast | Predicted future conditions. |
Radar rainfall is an estimate over an area. A forecast value is not evidence that rain fell. Polling a provider successfully does not refresh the observation timestamp inside its response.
Choose by purpose
| Source | Useful for | Check before relying on it |
|---|---|---|
| Tempest, Ecowitt, Davis | Readings from your own weather hardware | Installed sensors, local reception, observation age |
| Ambient Weather, Netatmo, La Crosse | Your station’s observations through its cloud | Hardware modules, credentials, cloud availability |
| Home Assistant passthrough | Existing HA weather and soil sensors | Entity mapping, units, source timestamp, precipitation interval |
| NWS observations | Measured conditions from an official station | Distance and freshness; a fresh temperature does not guarantee fresh wind |
| Synoptic Data | Observations from another physical station | Selected station, coverage, token, available fields |
| NOAA MRMS | Radar rainfall estimates in supported US coverage | Product age and accumulation period |
| Open-Meteo, NWS forecasts, MET Norway, Pirate Weather, OpenWeather, WeatherKit | Forecasts and modeled fields exposed by the adapter | Model coverage, forecast age, missing fields, provider requirements |
Near-real-time wind
Prefer a fresh on-site anemometer where available. NWS wind is measured, but its station may be distant or its wind observation stale. Open-Meteo wind is modeled.
Inspect the field’s source and observation time in LocalSky. A selected primary does not win when it has no eligible reading; the configured chain can fall through to another source. Decide whether a modeled fallback is suitable for your use case.
Rain needs its interval
Current rain intensity, rain in the preceding minute, today’s total, and forecast rain are different inputs. Do not interchange them based on a similar entity name.
For HA’s local WeatherFlow precipitation sensor, choose Rain last minute (accumulate today). LocalSky accumulates received intervals into daily totals; it cannot recover intervals missed while disconnected.
Measured rain history and forecast rain stay distinct in watering explanations and the API.
The current sky
The weather card and Home Assistant show one judgement of the sky, built from whatever your sources report:
- Rain, snow, hail and thunderstorms appear only when a gauge, radar or lightning sensor observes them. A forecast’s chance of storms is not a storm.
- Day and night follow the sun at your configured location, so an overcast morning reads as a cloudy day rather than a night.
- In daylight, a light sensor’s reading is compared with clear-sky sunlight for the sun’s height. At night, or without a light sensor, sky cover comes from a nearby station’s cloud report, then a provider’s current cloud cover, then the forecast hour.
- Fog comes from reported visibility.
- When nothing reports the sky, the card says so instead of guessing.
Forecast tracks
The main forecast serves the app and watering decisions. Extra models provide separate comparison and integration tracks. An NBM track, for example, remains a forecast, including when accessed through Open-Meteo.