High-resolution models and hourly updates for North America
GFS and HRRR models now integrated
NOAA NCEP provides one of the best weather models for North America. Their GFS and HRRR models are now available at Open-Meteo!
With the dedicated Open-Meteo GFS API you can access their forecast easily. No API key is required and it is free for non-commercial use.
A large variety of weather variables is available like temperature, humidity, wind, clouds, solar radiation, soil conditions and many more. Not only on surface, but also in the upper-air on 28 atmospheric pressure levels. This is great to study winds or cloud layers in the atmosphere.
Global Forecast System (GFS)
The GFS weather model is one of the oldest operational weather models starting in 1980 and a pioneer of open-data. With constant updates, forecast quality steadily increased. The latest upgrade in 2021 now features global 0.25° (~25 km) resolution with hourly data for 16 days of forecast.
GFS updates every 6 hours with initialisation data from 0:00, 6:00, 12:00 and 18:00 GMT. After current measurements, satellite, radar, ballon and airplane observations have been collected, the first forecast hour is available at around 3:30 GMT on the NOAA NCEP open data server. Consecutive forecast hours follow every couple of minutes and the last forecast hour to complete the 16 day forecast, arrives at 5:15 GMT.
The Open-Meteo API is tightly coupled with this update schedule and downloads data while being published. Afterwards it only takes 10 minutes until the latest GFS forecast update is seamlessly integrated into the Open-Meteo APIs.
High Resolution Rapid Refresh (HRRR) model
Weather for the next days is predominately determent by large scale weather patterns and GFS does a great job of predicting pressure systems, frontal systems or large scale precipitation.
“Hyperlocal” weather on the other hand is effected by convection with local showers and thunderstorms. Local vegetation, soil type or steep mountainous terrain changes convection and wind flow drastically.
A global model with 0.25° (~25 km) struggles to cope with local conditions. Over the years, parameterisations were developed to anticipate for many effects, but ultimately higher resolutions are required.
The HRRR model with 3 km resolution solves those short-comings. Not only with more finer resolution, but also with updates every hour. This greatly improves forecast for the next hours, because the latest radar observations help to forecast thunderstorms.
The first HRRR forecast hour is available 50 minutes after initialisation time and takes 30 minutes to calculate 18 hours of forecast. Every 6 hours forecasts with up to 48 hours can be downloaded.
Unfortunately, HRRR is only available for North-America and primarily covers US conus.
Calculating global and local weather models requires immense CPU power. Only High-Performance-Cluster with thousands of CPUs cores, specialised networking and fast storage can calculate weather models for real-time weather applications. With increasing computational resources, higher resolutions, larger areas and extended forecast times become possible.
The combination of local and global domains is a good compromise to predict large scale weather patterns and more detailed local forecasts.
Open-Meteo GFS API
The Open-Meteo GFS API transparently combines the global GFS model with local HRRR updates. Of course HRRR updates can only be applied to locations in North-America.
All weather model updates are integrated into a continuous time-series and accessing data from the past days, weeks or months, just works.
Compared to the “default weather forecast API“ there are some differences in weather variables. For example: Diffuse and shortwave radiation is calculated from solar separation models instead of being directly available. Also the “weather condition codes” are not yet compatible with the WMO weather codes. There is still work to do, but otherwise the API is in great shape and forecasts for the US significantly improved.
In the next weeks, more weather models will follow. For example MeteoFrance Arome. As every weather model is different, the challenge is maintaining the highest precision for each weather model. The ultimate goal is
One weather forecast API, that automatically provides the best weather forecast as a combination from many different weather models
Many specific APIs that offer the untainted and consistent forecasts from each national weather provider like NOAA NCEP, ECMWF, DWD or MeteoFrance.
With more and more weather models, this will be a great playground to train Machine-Learning algorithms with multiple data-sources and produce better forecasts.
Make sure to subscribe to this newsletter to get latest updates!