Research
Our state-of-the-art model provides 10-day weather forecasts with unprecedented accuracy in less than a minute.
Weather affects us all, in different ways. It can dictate how we dress in the morning, provide us with green energy and, in the worst cases, create storms that can devastate communities. In a world of increasingly extreme weather, fast and accurate forecasts have never been more important.
In a paper published in Science, we introduce GraphCast, a cutting-edge AI model capable of making medium-range weather forecasts with unprecedented accuracy. GraphCast predicts weather conditions up to 10 days in advance more accurately and much faster than the industry’s gold standard weather simulation system: High Resolution Forecast (HRES), produced by the European Center for medium-range weather forecasts (ECMWF).
GraphCast can also provide earlier warnings of extreme weather events. It can predict the path of cyclones with high accuracy in the future, identify atmospheric rivers associated with flood risks and predict the occurrence of extreme temperatures. This capability has the potential to save lives through better preparedness.
GraphCast takes a significant step forward in AI for weather forecasting, delivering more accurate and efficient forecasts and leading the way to support decision-making critical to the needs of our industries and societies. And by open source model code for GraphCast, we enable scientists and forecasters around the world to benefit billions of people in their daily lives. GraphCast is already used by weather agencies, including ECMWF, which is running a live experiment of the forecasts of our model on its website.
A selection of 10-day GraphCast forecasts showing specific humidity at 700 hectopascals (about 3 km above the surface), surface temperature and surface wind speed.
The challenge of global weather forecasting
Weather forecasting is one of the oldest and most difficult scientific activities. Medium-term forecasts are important to support key decision-making in every sector, from renewable energy to event logistics, but they are difficult to achieve accurately and efficiently.
Forecasts are typically based on numerical weather prediction (NWP), which begins with carefully defined physical equations, which are then translated into computer algorithms running on supercomputers. Although this traditional approach has been a triumph of science and engineering, designing the equations and algorithms is time-consuming and requires extensive expertise, as well as expensive computational resources to make accurate predictions.
Deep learning offers a different approach: using data rather than physical equations to create a weather forecast system. GraphCast draws on decades of historical weather data to learn a model of the cause-and-effect relationships that govern the evolution of Earth’s weather, from present to future.
Fundamentally, GraphCast and traditional approaches go hand in hand: we trained GraphCast on four decades of weather reanalysis data, from the ECMWF ERA5 dataset. This treasure trove is based on historical weather observations such as satellite images, radar and weather stations using traditional numerical weather prediction to “fill in the blanks” where observations are incomplete, to reconstruct a rich record of the weather world history.
GraphCast: an AI model for weather forecasting
GraphCast is a weather forecasting system based on machine learning and graph neural networks (GNN), which is a particularly useful architecture for processing spatially structured data.
GraphCast forecasts at high resolution of 0.25 degrees longitude/latitude (28 km x 28 km at the equator). That’s over a million grid points covering the entire surface of the Earth. At each grid point, the model predicts five Earth surface variables – including temperature, wind speed and direction, and average sea level pressure – and six atmospheric variables at each of 37 levels. altitude, including specific humidity, wind speed and direction, and temperature.
Although training GraphCast was computationally intensive, the resulting forecasting model is very efficient. Making a 10-day forecast with GraphCast takes less than a minute on a single Google TPU v4 machine. For comparison, a 10-day forecast using a conventional approach, such as HRES, can take hours of computing time in a supercomputer with hundreds of machines.
In a comprehensive performance evaluation against the benchmark deterministic system, HRES, GraphCast provided more accurate predictions on over 90% of 1,380 test variables and forecast lead times (see our Scientific article for more details). When we limited the evaluation to the troposphere, the region of the atmosphere between 6 and 20 kilometers altitude closest to the Earth’s surface where accurate forecasts are most important, our model outperformed HRES on 99.7% of test variables for future weather conditions.
For inputs, GraphCast only requires two sets of data: the weather status 6 hours ago and the current weather status. The model then predicts the weather in 6 hours. This process can then be advanced in 6 hour increments to provide peak forecasts up to 10 days in advance.
Better warnings for extreme weather events
Our analyzes found that GraphCast can also identify severe weather events earlier than traditional forecast models, even though it was not trained to look for them. This is a great example of how GraphCast could help prepare to save lives and reduce the impact of storms and extreme weather on communities.
By applying a simple cyclone tracker directly to GraphCast forecasts, we could predict cyclone movement more accurately than the HRES model. In September, a live version of our publicly available GraphCast model, deployed on the ECMWF website, accurately predicted about nine days in advance that Hurricane Lee would make landfall in Nova Scotia. In contrast, traditional forecasts had greater variability in location and timing of landfall and only predicted Nova Scotia about six days in advance.
GraphCast can also characterize atmospheric rivers – narrow regions of the atmosphere that transfer most water vapor out of the tropics. The intensity of an atmospheric river can indicate whether it will bring beneficial rain or a flood-causing deluge. GraphCast forecasts can help characterize atmospheric rivers, which could help plan emergency responses in collaboration with AI models to predict floods.
Finally, forecasting extreme temperatures is of increasing importance in a warming world. GraphCast can characterize when heat is expected to exceed historical maximum temperatures for a given location on Earth. This is particularly useful for anticipating heat waves, increasingly frequent disruptive and dangerous events.
Severe Event Prediction: How GraphCast and HRES Compare.
Left: cyclone tracking performance. As the lead time for forecasting cyclone movements increases, GraphCast maintains greater accuracy than HRES.
Right: atmospheric prediction of rivers. GraphCast’s prediction errors are significantly lower than HRES’s for their entire 10-day predictions
The future of AI for weather
GraphCast is now the world’s most accurate 10-day global weather forecast system and can predict extreme weather events further into the future than was previously possible. As weather conditions evolve in a changing climate, GraphCast will evolve and improve as better quality data becomes available.
To make AI-based weather forecasting more accessible, we have open source the code of our model. ECMWF is already experiment with GraphCast’s 10-day forecast and we’re excited to see the possibilities it opens up for researchers – from tailoring the model to particular weather phenomena to optimizing it for different parts of the world.
GraphCast joins other leading weather forecasting systems from Google DeepMind and Google Research, including a regional system Nowcasting model which produces forecasts up to 90 minutes in advance, and MetNet-3a regional weather forecast model already operational in the United States and Europe that produces more accurate 24-hour forecasts than any other system.
The pioneering use of AI in weather forecasting will benefit billions of people in their daily lives. But our broader research isn’t just about anticipating weather patterns: it’s also about understanding broader patterns in our climate. By developing new tools and accelerating research, we hope that AI can enable the global community to tackle our greatest environmental challenges.
