I am using the Historical Forecast API to retrieve past weather forecast data, and I have a question regarding the generation time of the forecast data.
If I download forecast data for January 2023, how can I determine when each forecast was originally generated?
Does the API provide metadata that indicates the forecast generation timestamp?
If not, does the Historical Forecast API provide forecast data that was issued 6 hours ago, 12 hours ago, or another specific time in the past?
I would appreciate any clarification on this. Thank you for your support!
Weather data is continuously assimilated from the latest model updates. In Europe and North America, high-resolution models update every 1 to 3 hours, so only the first 1 to 3 hours of each update are used. Global models update every 6 hours, using the first 6 hours of each run. Adding the forecast reference time for each timestep would help clarify this and is on our To-Do list. That said, there are many other improvements and additional weather models in the works.
If you're looking for past forecasts, you can use the "Previous Model Runs API," which provides lead time offsets of 24, 48, 72 hours, and beyond. More granular offsets of 6 or 12 hours aren’t available yet, but we're considering adding them.
I am working on a machine learning forecast model. Do we have the "forecast length" feature for the historical weather API? I need the previous 2-3 years of historical forecasts with a forecast length of at least 7 days for the Australian region.
I'm currently training a model that uses forecast weather variables as input. So my question is, are the historical forecast data downloaded from this API raw or have they been modified/improved through some process?
For example, is the forecast data provided for July 2 the output produced on July 1? I'm using the NOAA GFS weather model.
Basically what I need to know is if the data that I'm going to use for the model in production (let's say the next 24 hour forecast data from now) will have the same accuracy as the training data that this historical forecast API provides.
The Historical Forecast API combines all available runs. In case of GFS it is updated every 6 hours, therefore the lead-time offset is only a couple of hours. If you want to train for a fixed lead-time offset of 1 day, you can use the Previous Models Runs API and "previous_day_1". More information here: https://openmeteo.substack.com/p/weather-forecasts-from-previous-model-runs
The problem is that the previous model runs provide data from January this year and I need at least one year. I saw that JMA models does provide data from 2018, but the solar radiation variables are not available. Is there any way I can get a full year of historical forecast with temperature, humidity, wind speed and solar radiation data (more specifically, in Spain)?
Both, the regular "Forecast API" as well as "Historical Forecast API" contain the same weather models. If you will have to use both APIs to get historical and forecast data.
Thanks. Only the open-data ECMWF IFS 0.25° as well as AIFS 0.25° are available. Forecasts from the high resolution IFS 9 km cannot be offered due to license restrictions. However, IFS 9 km is available alongside ERA5 in the Historical Weather API with a delay of 2 days.
Hi Open-Meteo team,
I am using the Historical Forecast API to retrieve past weather forecast data, and I have a question regarding the generation time of the forecast data.
If I download forecast data for January 2023, how can I determine when each forecast was originally generated?
Does the API provide metadata that indicates the forecast generation timestamp?
If not, does the Historical Forecast API provide forecast data that was issued 6 hours ago, 12 hours ago, or another specific time in the past?
I would appreciate any clarification on this. Thank you for your support!
Hi Jiyun!
Weather data is continuously assimilated from the latest model updates. In Europe and North America, high-resolution models update every 1 to 3 hours, so only the first 1 to 3 hours of each update are used. Global models update every 6 hours, using the first 6 hours of each run. Adding the forecast reference time for each timestep would help clarify this and is on our To-Do list. That said, there are many other improvements and additional weather models in the works.
If you're looking for past forecasts, you can use the "Previous Model Runs API," which provides lead time offsets of 24, 48, 72 hours, and beyond. More granular offsets of 6 or 12 hours aren’t available yet, but we're considering adding them.
Hi, great work! Keep it up!
I am working on a machine learning forecast model. Do we have the "forecast length" feature for the historical weather API? I need the previous 2-3 years of historical forecasts with a forecast length of at least 7 days for the Australian region.
Hi, you can get forecasts from previous days through the Previous Runs API. However only data starting in January 2024 is available: https://open-meteo.com/en/docs/previous-runs-api
Hi!
Thanks for creating and improving this great API!
I'm currently training a model that uses forecast weather variables as input. So my question is, are the historical forecast data downloaded from this API raw or have they been modified/improved through some process?
For example, is the forecast data provided for July 2 the output produced on July 1? I'm using the NOAA GFS weather model.
Basically what I need to know is if the data that I'm going to use for the model in production (let's say the next 24 hour forecast data from now) will have the same accuracy as the training data that this historical forecast API provides.
The Historical Forecast API combines all available runs. In case of GFS it is updated every 6 hours, therefore the lead-time offset is only a couple of hours. If you want to train for a fixed lead-time offset of 1 day, you can use the Previous Models Runs API and "previous_day_1". More information here: https://openmeteo.substack.com/p/weather-forecasts-from-previous-model-runs
Thanks for your reply!
The problem is that the previous model runs provide data from January this year and I need at least one year. I saw that JMA models does provide data from 2018, but the solar radiation variables are not available. Is there any way I can get a full year of historical forecast with temperature, humidity, wind speed and solar radiation data (more specifically, in Spain)?
helloo, im in the same situation as you, did you manage to fix it?
Hi, thank you for the continuous improvments.
Is that possible to get historical forecasts with a horizon (or forecast days) larger than 3 hours ?
Sure. You can just use the regular weather Forecast API. Up to 16 days of forecast are available: https://open-meteo.com/en/docs
thanks ! But weather forecast api can deliver recent past days forecasts only. How can we get historical forecasts from several years ago?
Both, the regular "Forecast API" as well as "Historical Forecast API" contain the same weather models. If you will have to use both APIs to get historical and forecast data.
I’d love to learn more about how the historical data is stored. Have you considered Zarr?
This is an amazing work that will definitely make an easier pipeline for research. Will the dataset contain ECMWF forecast as well?
Thanks. Only the open-data ECMWF IFS 0.25° as well as AIFS 0.25° are available. Forecasts from the high resolution IFS 9 km cannot be offered due to license restrictions. However, IFS 9 km is available alongside ERA5 in the Historical Weather API with a delay of 2 days.