GraphCast: An AI model for faster and more accurate global weather forecasting

GraphCast: An AI model for faster and more accurate global weather forecasting



Remy Lam on behalf of the GraphCast team

Our advanced model provides 10-day weather forecasts with unprecedented accuracy in less than one minute

Weather affects us all, in big and small ways. It can dictate how we get dressed in the morning, provide us with green energy, and in the worst cases, create storms that can destroy communities. In a world of increasingly severe weather, fast and accurate forecasts are more important than ever.

In a paper published in the journal Science, we present GraphCast, a state-of-the-art AI model capable of providing medium-range weather forecasts with unprecedented accuracy. GraphCast forecasts weather conditions up to 10 days in advance more accurately and much faster than the industry's gold standard weather simulation system – High Resolution Forecasts (HRES), produced by the European Center for Medium-Range Weather Forecasts (ECMWF).

GraphCast can also provide early warnings about severe weather events. It can predict future hurricane paths with great accuracy, identify atmospheric rivers associated with flood risk, and predict the onset of extreme temperatures. This capability has the potential to save lives by increasing preparedness.

GraphCast takes an important step forward in artificial intelligence for weather forecasting, delivering more accurate and efficient forecasts, and opening paths to support decision-making critical to the needs of our industries and societies. By open sourcing GraphCast's sample code, we enable scientists and forecasters around the world to benefit billions of people in their daily lives. GraphCast is already being used by meteorological agencies, including ECMWF, which is running a live trial of our model predictions on its website.

A selection of 10-day GraphCast forecasts show 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 challenging scientific endeavors. Medium-range forecasts are important to support key decision-making across sectors, from renewable energy to event logistics, but they are difficult to do accurately and efficiently.

Forecasts are typically based on numerical weather prediction (NWP), which starts with carefully defined physics equations, which are then translated into computer algorithms running on supercomputers. While this traditional approach has been a triumph of science and engineering, designing equations and algorithms is time-consuming and requires deep expertise, as well as expensive computational resources to make accurate predictions.

Deep learning offers a different approach: using data instead of physical equations to create a weather forecasting system. GraphCast is trained on decades of historical weather data to learn a model of the cause-and-effect relationships that govern how Earth's weather evolves, from the present to the future.

Importantly, GraphCast and traditional methods go hand in hand: we trained GraphCast on four decades of weather reanalysis data, from ECMWF's ERA5 dataset. This trove is based on historical weather observations such as satellite images, radar and weather stations that use traditional numerical weather prediction to “fill in the blanks” where observations are incomplete, to reconstruct a rich record of global historical weather.

GraphCast: An artificial intelligence model for weather forecasting

GraphCast is a weather forecasting system based on machine learning and graph neural networks (GNNs), an architecture particularly useful for processing spatially structured data.

GraphCast makes forecasts with a high accuracy of 0.25 degrees longitude/latitude (28 km x 28 km at the equator). This means more than a million grid points covering the entire surface of the Earth. At each grid point, the model predicts five land surface variables—including temperature, wind speed and direction, and mean sea level pressure—and six atmospheric variables at each of 37 elevation levels, including specific humidity. Wind speed and direction. Temperature.

Although training GraphCast was computationally intensive, the resulting prediction model is highly efficient. Making a 10-day forecast with GraphCast takes less than a minute on a single Google TPU v4 device. By comparison, a 10-day forecast using a traditional approach, such as HRES, can take hours of computation in a supercomputer with hundreds of machines.

In a comprehensive evaluation of performance against the gold standard deterministic system, HRES, GraphCast provided more accurate predictions for over 90% of 1,380 test variables and prediction intervals (see our Science paper for details). When we restricted the evaluation to the troposphere, the 6-20 km region of the atmosphere closest to the Earth's surface where accurate prediction is most important, our model outperformed HRES on 99.7% of the test variables for future weather.

For input, GraphCast requires only two data sets: weather 6 hours ago, and current weather. The model then predicts the weather 6 hours in the future. This process can then be performed in 6-hour increments to provide up-to-the-minute forecasts up to 10 days in advance.

Better warnings for severe weather events

Our analyzes revealed that GraphCast can also identify severe weather events earlier than traditional forecast models, even though it is not trained to look for them. This is a prime example of how GraphCast can help prepare to save lives and reduce the impact of storms and severe weather on communities.

By applying a simple hurricane tracker directly to GraphCast forecasts, we can predict hurricane movement more accurately than the HRES model. In September, a live version of the publicly available GraphCast model, posted on the ECMWF website, accurately predicted about nine days before Hurricane Lee would make landfall in Nova Scotia. In contrast, conventional forecasts had greater variability in where and when landfall occurred, and were only recorded in Nova Scotia about six days earlier.

GraphCast can also describe atmospheric rivers — narrow regions of the atmosphere that transport most of the 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 may help plan emergency responses in conjunction with artificial intelligence models for flood forecasting.

Finally, forecasting extreme temperatures is increasingly important in our warming world. GraphCast can pinpoint when heat is set to rise above the highest historical temperatures for any given location on Earth. This is particularly useful in forecasting heatwaves and disruptive and dangerous events that are becoming increasingly common.

Critical Event Forecasting – How GraphCast and HRES compare.

Left: Hurricane tracking views. As the lead time required to forecast hurricane movements increases, GraphCast maintains greater accuracy than HRES.

Right: Atmospheric river forecast. GraphCast's forecast errors are significantly lower than those of HRES across its 10-day forecast

The future of weather artificial intelligence

GraphCast is now the most accurate global 10-day weather forecast system in the world, and can predict more extreme weather events in the future than was previously possible. As weather patterns evolve in a changing climate, GraphCast will evolve and improve as higher quality data becomes available.

To make AI weather forecasting more accessible, we've open sourced our model code. ECMWF is already piloting 10-day GraphCast forecasts, and we are excited to see the possibilities it opens up for researchers – from tailoring the model for specific climate phenomena to optimizing it for different parts of the world.

GraphCast joins other cutting-edge weather forecasting systems from Google DeepMind and Google Research, including a regional nowcasting model that produces forecasts up to 90 minutes in advance, and MetNet-3, a regional weather forecasting model already in operation across the United States. Europe produces more accurate 24-hour forecasts than any other system.

Pioneering the use of artificial intelligence in weather forecasting will benefit billions of people in their daily lives. But our broader research is not just about predicting the weather, it's about understanding the broader patterns of our climate. By developing new tools and accelerating research, we hope that AI can empower the global community to address our greatest environmental challenges.

You may also like...

Leave a Reply

Your email address will not be published. Required fields are marked *