FourCastNet and GraphCast Now Available in the Weather API

We are happy to announce the integration of the FourCastNet and GraphCast AI models into the Meteomatics Weather API, complementing our suite of weather forecasting solutions.

A Complete Offer With All Kinds of Weather Data

With the inclusion of FourCastNet and GraphCast AI models, our Weather API now provides even more weather forecasting solutions.

While our focus remains on numerical weather models, such as our in-house developed 1-kilometer resolution models, the integration of AI-based forecasts supports our goal of serving as a one-stop weather data shop, meeting the varied needs of our customers.

NVIDIA’s FourCastNet

The addition of FourCastNet is the outcome of a collaborative effort with NVIDIA. Short for “Fourier Forecasting Neural Network”, NVIDIA’s FourCastNet is GPU-accelerated, powered by the Fourier Neural Operator, and trained on 10 TB of Earth system data. It provides short to medium-range global predictions (20 days) at the resolution of ca. 25 km, and generates 7-day forecasts in less than 2 seconds.

Google’s GraphCast

GraphCast is an open-source project developed by Google DeepMind, aiming to make advanced weather forecasting more accessible. It uses machine learning and Graph Neural Networks (GNNs), and, like FourCastNet, provides medium-range weather forecasts (20 days) with a resolution of ca. 25 km. GraphCast generates a 10-day forecast in less than a minute on a single Google TPU v4 machine.

Timing Is King: AI Models Give You a Time Advantage

When time is money, timing is king. While high-resolution numerical weather models, such as EURO1k, are the most reliable for short-term forecasts, AI-based weather forecasting can give a competitive edge to industries looking further in time, notably energy trading.

FourCastNet and GraphCast offer 20-day forecasts four times daily, while ECMWF, the global weather model standard, provides 10-day forecasts only twice daily.

Additionally, ECMWF takes about 50 minutes to complete a full model run for a 10-day forecast. In contrast, AI-based models only need ECMWF's first file to start their run, allowing them to produce forecasts in just 15 minutes, which gives users a time advantage of around 35 minutes.

Enhanced Forecasts With Meteomatics 90-meter Downscaling

The output of AI-based forecasting models will also benefit from Meteomatics' signature 90-meter downscaling feature. This advanced downscaling capability enhances the spatial resolution of weather data, providing users with highly localized forecasts tailored to their specific geographical areas of interest.

You're already a Weather API user

If you already have a Weather API account and would like to know more about the AI-based weather models, your dedicated client manager will be happy to support you!

You're not yet a Weather API user

If you’re new here, why not try our Free Trial? With a Free Trial account, you receive access to all the weather sources and models available in the API, during 14 days.