Advances in Modeling Atmospheric Observational Variables Using DeepMind and Graph Neural Networks

Today, there was peer-reviewed evidence that proved that promise. In a paper that was published today in Science, researchers from DeepMind showed that the model it used did better than the forecasts from the European Centre for Medium-Range Weather Forecasting. Better yet, the DeepMind model could be run on a laptop and spit out a forecast in under a minute, while the conventional models require a giant supercomputer.

Standard weather simulations try to replicate the atmosphere’s physics. They’ve gotten better over the years, thanks to better math and by taking in fine-grained weather observations from growing armadas of sensors and satellites. They are also very cumbersome. It can take hours forforecasts at major weather centers to be calculated.

Battaglia was already leading a team focused on applying AI systems called graph neural networks, or GNNs, to model the behavior of fluids, a classic physics challenge that can describe the movement of liquids and gases. It seemed very logical to tap GNNs since weather prediction is at its core about modeling the flow of molecule. While training these systems is heavy-duty, requiring hundreds of specialized graphics processing units, or GPUs, to crunch tremendous amounts of data, the final system is ultimately lightweight, allowing forecasts to be generated quickly with minimal computer power.

According to the computer scientist at University of California, Los Angeles, GraphCast is currently leading the race. The model is described1 in Science on 14 November.

“In the troposphere, which is the part of the atmosphere closest to the surface that affects us all the most, GraphCast outperforms HRES on more than 99% of the 12,00 measurements that we’ve done,” says computer scientist Remi Lam at DeepMind in London. Across all levels of the atmosphere, the model outperformed HRES on 90% of weather predictions.

GraphCast projected the state of 5 weather variables, which were close to the Earth’s surface, as well as 6 atmospheric variables like windspeed, further from the earth’s surface.

It also proved useful in predicting severe weather events, such as the paths taken by tropical cyclones, and extreme heat and cold episodes, says Chantry.

Machine Learning for Prediction of Weather: An Empirical Study of GraphCast vs. Other Models in Chantry’s Study

Chantry noted that GraphCast performed better than other models in the study because of its evaluation by some metrics, but future assessments of performance using other metrics could lead to a different result.

Rather than entirely replacing conventional approaches, machine-learning models, which are still experimental, could boost particular types of weather prediction that standard approaches aren’t good at, says Chantry — such as forecasting rainfall that will hit the ground within a few hours.

AI models also run the risk of amplifying biases in their training data and require a lot of energy for training, although they consume less energy than NWP models, says Grover.