⚡️ Features | Open-Meteo.com

Global weather data from forecast to archive — free, no sign-up, one JSON API.
Open-Meteo integrates numerical weather prediction models from over 15 national weather services, including ECMWF, DWD, NOAA, Météo-France, JMA, KMA, KNMI, DMI, MeteoSwiss, UK Met Office, BOM, CMA, and GeoSphere Austria.
The best_match option automatically selects the most suitable high-resolution model for any location. Individual models can also be selected directly, enabling model-specific workflows and comparisons.
The API selects the highest-resolution model available for each location. In Central Europe, France, Switzerland, the UK, and North America, high-resolution models at 1–2 km are used for the first few days of the forecast. Global models at 9–11 km cover the remainder of the forecast horizon worldwide.
All data is optimised to the requested coordinates and normalised to a consistent hourly time step, regardless of the native model resolution or output interval.
Most global models are updated every 6 hours. High-resolution regional models such as ICON-D2 and AROME update every 1–3 hours. Each update ingests the latest observations from weather stations, radiosondes, aircraft, radar, and satellites.
15-minutely data is available for Central Europe and North America, interpolated from hourly model output elsewhere.
Open-Meteo provides four distinct historical data products, each suited to different use cases:
Beyond standard surface variables, the APIs provide: solar radiation (including global tilted irradiance for arbitrary panel orientation), wind at multiple altitudes (10 m to 300 m+), pressure levels (1000–10 hPa), soil moisture and temperature at multiple depths, ocean waves, air quality with pollen, and ensemble probabilistic forecasts.
The Seasonal Forecast API provides ECMWF SEAS5 forecasts out to 9 months and sub-seasonal ECMWF EC46 forecasts out to 6 weeks, both at 36 km resolution with 51 ensemble members.
The ensemble spread represents uncertainty at these extended horizons. Data are not bias-corrected and should be interpreted as probabilistic area guidance — whether the coming weeks or months are likely to be warmer, colder, wetter, or drier than the climatological average.
Training weather post-processing models, backtesting energy forecasts, and measuring weather model skill all require different types of historical data. Open-Meteo provides four complementary APIs covering distinct use cases.
Historical Weather API
ERA5 and ERA5-Land reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) covers the globe at 9–25 km resolution with hourly data from January 1940 to the present. Unlike weather station data, reanalysis provides spatially complete, consistent coverage with no missing values.
This is the standard dataset for climate baselines, anomaly detection, long-range correlations, and training models where decades of labelled examples are needed. The API delivers data lightning-fast — a multi-decade time series for a single location returns in under 100 ms.
Historical Forecast API
The Historical Forecast API archives the initial hours of each weather model update into a seamless, continuously updated time series — the same data structure as the live Forecast API. This makes it directly comparable to real operational forecasts.
This is the right dataset when you need weather model output that is consistent with the live forecast for training bias-correction and post-processing models: the historical archive and real-time forecast use the same variables, units, and time alignment. Data is available for all models supported by the Forecast API.
The Previous Runs API provides time-series data at a fixed lead-time offset from 1 to 7 days. Instead of returning individual runs in full, it returns a continuous time series where each value was forecast exactly N days in advance — e.g. a `_previous_day2` series contains only values taken from the run initialised 48 hours before valid time. Data is available from January 2024.
This makes it straightforward to compute systematic forecast error (bias) at each lead time, compare model skill across offsets, and build training datasets where the forecast horizon is a controlled variable.
The Single Runs API provides access to archived individual model runs, selected by their exact initialisation time via the run= parameter. Each run covers the full forecast horizon, typically 7–10 days. ECMWF IFS HRES at native 9 km resolution is available from March 2024 via hindcasts.
This is the correct dataset for: training weather post-processing models without look-ahead bias (each training sample is a single run paired with verifying observations); reconstructing the exact forecast available at a past decision time for energy trading or grid operation backtests; and run-level model intercomparison across providers.
All APIs share the same JSON format and parameter conventions. Switching between them requires only a change of hostname or endpoint.
Hourly and daily weather forecasts for any global location. Returns temperature, wind, precipitation, solar radiation, soil variables, and more. Supports multiple models or automatic best-match selection. Past days parameter provides seamless access to recent history without switching endpoints.
ERA5 and ERA5-Land data at hourly resolution from January 1940 to present. Global coverage at 9–25 km with no missing data. Used for climate analysis, anomaly detection, and as ground truth for training machine learning models.
Seamless archive of weather forecast model output, built by continuously storing the initial-hours data from each model update. The format is identical to the live Forecast API, making it the standard dataset for training bias-correction and post-processing pipelines.
Returns a continuous time series where every value was forecast at a fixed lead-time offset of 1–7 days. Available from January 2024. Suited for systematic bias analysis, lead-time-stratified ML training, an