A Global Standard is Beating a DeepMind Artificial Intelligence Weather Forecaster

DeepMind: Predicting Climate and Molecular Atmosphere with Large Scale Artificial Intelligence Models (Extended Abstract)

The new evidence of that promise has been shared by the company. In a paper published today in Science, DeepMind researchers report that its model bested forecasts from the European Centre for Medium-Range Weather Forecasting (ECMWF), a global giant of weather prediction, across 90 percent of more than 1,300 atmospheric variables such as humidity and temperature. 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.

Predicting the weather is incredibly difficult and takes a lot of energy. The standard approach to forecasting weather is called numerical weather prediction. The physical models crunch weather data from all over the world using the power of supercomputers. The calculations accurately map out how heat, air and water move through the atmosphere, but it’s expensive and energy-intensive to run.

Battaglia and his team are applying artificial intelligence to model the behavior of fluids in order to better understand the movement of liquids and gases. It seemed intuitive to tap the GNNs given the core role of weather prediction in modeling the flow of Molecules. The system is lightweight, and allows forecasts to be generated quickly with minimal computer power, because it requires hundreds of specialized graphics processing units to crunch huge amounts of data.

“GraphCast currently is leading the race amongst the AI models,” says computer scientist Aditya Grover at University of California, Los Angeles. The model is described1 in Science on 14 November.

The computer scientist at DeepMind in London says GraphCast performs better than HRES in the atmosphere that is closest to the surface. The model was better at predicting weather on 90% of the time.

The state of atmospheric variables such as wind speed, which are further away from the Earth’s surface, was predicted by GraphCast.

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.

Chantry GraphCast and Machine Learning: Implications for Forecasting Rainfall and other Weather Prediction Methods in the Future

Chantry notes that although GraphCast’s performance was superior to other models in this study, based on its evaluation by certain metrics, future assessments of its performance using other metrics could lead to slightly different results.

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.

Although they consume less energy than NWP models, artificial intelligence models run the risk of having biases and require a lot of energy for training.