Historical Forecast API

The weather data precisely aligns with the weather forecast API, created by continuously integrating weather forecast model data. Each update from the weather models' initial hours is compiled into a seamless time series. This extensive dataset is ideal for training machine learning models and combining them with forecast data to generate optimised predictions.

Weather models are initialized using data from weather stations, satellites, radar, airplanes, soundings, and buoys. With high update frequencies of 1, 3, or 6 hours, the resulting time series is nearly as accurate as direct measurements and offers global coverage. In regions like North America and Central Europe, the difference from local weather stations is minimal. However, for precise values such as precipitation, local measurements are preferable when available.

The Historical Forecast API archives comprehensive data, including atmospheric pressure levels, from all accessible weather forecast models. Depending on the model and public archive availability, data is available starting from 2021 or 2022.

The default Best Match option selects the most suitable high-resolution weather models for any global location, though users can also manually specify the weather model. Open-Meteo utilises the following weather forecast models:

Open-Meteo offers four distinct historical weather datasets, each suited to different use cases. Only a small fraction of the Earth's surface has reliable, continuous weather station coverage; all four datasets use numerical weather models to fill that gap globally.

• Historical Weather API: ERA5 reanalysis at 0.25° (~25 km) from 1940, ERA5-Land at 0.1° (~9 km) from 1950, and ECMWF IFS analysis at 9 km from 2017. Optimised for long-term consistency rather than day-to-day accuracy — the right choice for climate trend analysis.
• Historical Forecast API: A continuous hourly timeseries built by stitching the first hours of each successive model run. Closely tracks actual conditions because each run is initialised from real measurements. Coverage starts around 2021. Not suitable for long time series due to model version changes over time.
• Previous Runs API: Archives the same high-resolution models as the Historical Forecast API, but provides each variable at a fixed lead-time offset: 1, 2, 3, up to 7 days ahead. Useful for evaluating forecast skill at specific lead times. Data starts from January 2024 (GFS from March 2021, JMA from 2018).
• Single Runs API: Retrieves the complete forecast horizon of any individual model run, selected by initialisation time using the run= parameter (e.g. run=2025-09-01T00:00). Unlike the Historical Forecast API — which stitches runs into a continuous timeseries — the Single Runs API preserves the original run structure. ECMWF IFS HRES at 9 km is archived from March 2024; all other models from September 2025.

Choosing the Right Dataset:

• For multi-decade climate analysis, use the Historical Weather API (ERA5 from 1940).
• For the most accurate representation of past conditions over the last few years, use the Historical Forecast API.
• To assess how forecast accuracy degrades at longer lead times, use the Previous Runs API.
• To retrieve the full output of a specific model run — for example to reproduce a forecast issued on a given date — use the Single Runs API.
• For training machine learning models on consistent weather model output, the Historical Forecast API and Single Runs API both provide data directly from the operational model runs.

As the API is identical to the Forecast API, please refer to the Weather Forecast API documentation for all available variables and parameters. The only notable difference is the API host "historical-forecast-api.open-meteo.com" as historical data is moved to a different set of servers with access to a large storage system.