Wednesday, July 24, 2024
HomeAI News & UpdatesGraphCast Revolutionizes AI Weather Forecasting

GraphCast Revolutionizes AI Weather Forecasting

Google DeepMind developers are adamant that The GraphCast AI model “marks a turning point in weather forecasting,” in a peer-reviewed paper published on Tuesday, November 14, 2023, in the journal Science.

For the first time, artificial intelligence outperformed traditional forecasting methods in predicting weather worldwide up to 10 days in advance. A comprehensive examination revealed that GraphCast was more accurate than the European Centre for Medium-Range Weather Forecasts’ leading conventional approach for projections three to ten days ahead. It exceeded the ECMWF product in 90% of the 1,380 variables tested, including temperature, pressure, wind speed and direction, and humidity at various levels of the atmosphere.

DeepMind’s claim that its system is the most accurate was backed up by Chantry. “We find GraphCast to be consistently more skillful than the other machine-learning models, Pangu-Weather from Huawei and FourCastNet from Nvidia, and on lots of scores, it is more accurate than our forecasting system,” he said in the interview with The Financial Times. GraphCast employs a machine-learning architecture known as graph neural network, which has learned about how weather systems develop and travel worldwide from more than 40 years of ECMWF data.

The ECMWF, an intergovernmental organization based in Reading in the UK, has been running live forecasts using AI models from Huawei, Nvidia, and DeepMind with its integrated forecasting system. According to Matthew Chantry, ECMWF’s machine-learning coordinator, AI systems in meteorology have advanced “far sooner and more impressively than we expected even two years ago.”


The inputs for GraphCast projections are the global atmospheric conditions at the current time and six hours ago, as compiled by ECMWF from global weather data. On a single Google TPU v4 cloud computer, GraphCast generates a 10-day forecast in under a minute.

In contrast to this data-derived “black box” approach, the traditional method used by ECMWF and the world’s national meteorological offices, known as numerical weather prediction, uses supercomputers to crunch equations based on scientific knowledge of atmospheric physics, which is a time-consuming and energy-intensive process.

Matthew Chantry said, “Once trained, GraphCast is tremendously cheap to operate,” adding, “We could be talking about 1,000 times less energy consumption. That is a remarkable improvement.”

DeepMind scientists cited Hurricane Lee in the North Atlantic in September as an example of a successful forecast. “GraphCast correctly predicted Lee’s landfall in Nova Scotia nine days before it occurred, compared to only six days for traditional approaches,” said Rémi Lam, principal author of the Science research. “That gave people three more days to prepare for its arrival.” However, AI did not outperform traditional physical models in predicting the sudden explosive intensification of Hurricane Otis near Mexico’s Pacific coast, which ravaged Acapulco on October 25 with little notice.

According to Chantry, the ECMWF’s next step would be to develop its own AI model and combine it with its numerical weather prediction system. “There is room to inject our understanding of physics into these machine-learning systems, which can seem like black boxes.”

The UK Met Office, the weather agency of the UK, announced a collaboration with the Alan Turing Institute, Britain’s AI research center, last month to construct its graph neural network for weather forecasting, which it will include in its existing supercomputer infrastructure.

The Met Office’s science director, Simon Vosper, emphasized the need to account for climate change in forecasting. “It is fair to question whether AI-based systems are able to pick up new extremes if these systems have only been ‘trained’ on previous weather conditions,” he added. “We aim to pull through the best that AI can offer while working with our traditional computer models based on the physics of the atmosphere,” Vosper said in a statement. “We believe that this blending of technologies will provide the most robust and detailed weather forecasts in an era of dramatic change.”


Editorial Staff
Editorial Staff
Editorial Staff at AI Surge is a dedicated team of experts led by Paul Robins, boasting a combined experience of over 7 years in Computer Science, AI, emerging technologies, and online publishing. Our commitment is to bring you authoritative insights into the forefront of artificial intelligence.


Please enter your comment!
Please enter your name here

Most Popular

Recent Comments