AI surpasses conventional weather forecasting methods – 11/14/2023 – Market

Artificial intelligence (AI) has convincingly outperformed conventional forecasting methods for the first time in predicting weather across the world for the ten days following the estimate.

The GraphCast AI model “marks a turning point in weather forecasting,” its Google DeepMind developers said in a peer-reviewed paper published in Science on Tuesday.

An extensive evaluation showed that GraphCast was more accurate than the world’s leading conventional system for forecasting three to ten days into the future, which is run by the European Center for Medium-Range Weather Forecasts (ECMWF).

It outperformed the ECMWF product by 90% of the 1,380 metrics used, which included temperature, pressure, wind speed and direction, and humidity at different levels of the atmosphere.

Matthew Chantry, machine learning coordinator at ECMWF, said AI systems in meteorology have progressed “much more quickly and more impressively than we expected even two years ago.”

ECMWF, an intergovernmental body based in Reading, UK, has been running live forecasts using AI models from Huawei and Nvidia, as well as DeepMind, alongside its own integrated forecasting system.

Chantry endorsed DeepMind’s claim that its system is the most accurate. “We found that GraphCast is consistently more skilled than the other machine learning models, Huawei’s Pangu-Weather and Nvidia’s FourCastNet, and in many ways more accurate than our own forecasting system,” he told the Financial Times .

GraphCast uses a machine learning architecture called a graph neural network, which has learned from more than 40 years of past ECMWF data about how weather systems develop and move around the globe.

The inputs to its forecasts are the states of the atmosphere around the world at the current time and six hours earlier, assembled by ECMWF from global weather observations. GraphCast produces a ten-day forecast in one minute on a single Google TPU v4 cloud computer.

In contrast to this data-derived “black box” approach, the conventional method used by ECMWF and the world’s national meteorological offices, known as numerical weather prediction, uses supercomputers to process equations based on scientific knowledge of atmospheric physics – a energy-intensive process that takes several hours.

“Once trained, GraphCast is extremely cheap to operate,” said Chantry. “We could be talking about a miraculous improvement in terms of energy consumption, about a thousand times cheaper.”

As an example of a successful forecast, DeepMind scientists mentioned Hurricane Lee in the North Atlantic in September. “GraphCast was able to correctly predict that Lee would hit Nova Scotia nine days before it happened, compared to just six days for traditional approaches,” said Rémi Lam, lead author of the Science paper. “This gave people three more days to prepare for his arrival.”

However, AI did no better than conventional physical models in predicting the sudden explosive intensification of Hurricane Otis off Mexico’s Pacific coast, which devastated Acapulco with little warning on October 25.

The next step for ECMWF would be to build its own AI model and analyze the combination with its numerical weather prediction system, Chantry said. “There is room to inject our understanding of physics into these machine learning systems, which can look like black boxes.”

The UK Met Office, the national meteorological service, announced last month a collaboration with the Alan Turing Institute, a British AI research center, to develop its own graph neural network for weather forecasting, which will be incorporated into its weather infrastructure. existing supercomputer.

Simon Vosper, director of science at the Met Office, highlighted the need to consider climate change in forecasting. “It’s fair to question whether AI-based systems are capable of identifying new extremes if these systems have only been ‘trained’ on previous climate conditions,” he said.

“Our goal is to leverage the best that AI can offer while working with our traditional physics-based computational models of the atmosphere,” Vosper added. “We believe this combination of technologies will provide the most robust and detailed weather forecasts in an era of dramatic change.”