It’s the race for reliable long-term weather forecasts | Science

The consequences of effective weather forecasting extend far beyond vacation planning. Having precise data over a long period, beyond the three days currently assumed by the most reliable estimates, would save lives and avoid economic losses, according to a study published in natural communications figure at 143 billion dollars per year (131.565 million euros). Technological giants like Google or IBM, in collaboration with NASA, and institutions from the EU and other continents, have joined the objectives of the United Nations plan improve early warning systems and develop tools to take advantage of advances in artificial intelligence in order to achieve reliable medium- and long-term forecasts. Understanding the sky is one of the scientific and technological challenges of our time.

Storms, floods, droughts, heatwaves, hurricanes and other extreme weather events represent billions in costs each year, most of which (63%) correspond to the assessment of human losses, according to the study of Nature led by New Zealand researchers. Understanding and anticipating these adverse events is essential and has become a key objective of the technology agenda.

Google DeepMind, the artificial intelligence company of the North American technology giant, published in Science a machine learning-based weather forecasting model to provide 10-day forecasts that are “better, faster and more accessible than existing approaches,” according to the study. The model, called GraphCast, outperformed traditional systems in 90% of cases tested.

Compared to current digital data analysis models that use expensive and complex computing resources, GraphCast uses machine learning trained on historical data to provide an accurate 10-day forecast in less than a minute. “We believe this marks a turning point in weather forecasting,” say the authors, led by DeepMind scientist Remi Lam.

In this race there is also IBM, in collaboration with NASA, with a proposal also for machine learning based on the founding model (trained with a wide spectrum of unlabeled data) geospatial of the technology company. This approach allows you to analyze millions of general data points to perform different tasks.

“Fundamental AI models using geospatial data (weather, sensors and satellite) can be a game-changer in helping us better understand, prepare for and address the many climate-related phenomena that impact the health of our planet. “in a way and at a speed never before seen,” says Alessandro Curioni, IBM vice president for Europe and Africa and director of the company’s research center in Zurich, Switzerland.

The program has already been used to analyze urban heat islands to reduce heat stress by up to three degrees Celsius and to plan a campaign to reforest 15 billion trees in Kenya over the next decade. The impact of extreme weather conditions on aircraft operations and infrastructure and a project to naturally restore forest masses to prevent flooding are also being explored, with the UK’s Science and Technology Facilities Council (STFC). There is already a pilot experiment in Glasgow.

Kate Royse, director of STFC’s Hartree Centre, said these models “enable smarter decisions based on accurate flood risk forecasting and management, which is essential for future city planning”.

“We are dramatically confronted with the accelerated effects of climate change. We need to do a better job of mitigating and preparing for these events. ‘AI could contribute to this,’ he warns in European research publication horizon professor at the Polytechnic University of Milan Andrea Castelletti, expert in natural resource management

Catelletti agrees with the IT giants’ approach since current prediction models are based on algorithms that allow large amounts of data to be analyzed without achieving optimal accuracy. “They still have weaknesses,” Castelletti admits, saying: “Artificial intelligence could solve them.”

“Existing climate models are not very suitable for certain extreme weather events. Heat waves in Europe, for example, are increasing much faster in the real world than models predict. It is important to predict extremes in order to be able to be alerted in time,” said Dim Coumou, climate expert at the University of Amsterdam (Netherlands).

CLINT, An EU-funded research project in which Spain is participating combines artificial intelligence with data from the European Copernicus satellite network to improve climate prediction. He European Center for Medium-Range Weather Forecasts, like IBM and Google, it also applies machine learning in its programs.

Another European project called XAIDA, and in which Spain also participates, attempts to understand the underlying causes of extreme weather events. “It’s about understanding the role of climate change in phenomena ranging from heat waves to droughts and extreme precipitation. We want to know the determining factors,” explains Coumou.

Undesirable and rare phenomena

One of the obstacles to achieving this accuracy is the information available to train artificial intelligence. Even though it appears that the information has been around for decades, it may not be relevant to understanding the most adverse events. “Extreme events are, by definition, rare. So you don’t always have a lot of observations. This is a big obstacle if we want to use artificial intelligence methods,” explains Coumou.

In this sense, the CLINT project aims for AI systems to be the ones that generate this data from historical information and allow machine learning to be trained to improve predictions.

Some initiatives focus on specific phenomena to make precise predictions. This is the case of CRUCIAL laboratory from the universities of Lancaster and Exeter, in the United Kingdom, which attempts to determine the number of hurricanes in the Atlantic in 2024.

“Changes in ocean temperatures, driven by climate change, mean that historical records of hurricanes are no longer a good guide to predicting future hurricanes,” he says. Kim Kaivantoprofessor of economics and member of the CRUCIAL initiative.

Likewise, researchers from Korea Institute of Civil Engineering and Construction Technology (KICT) have developed a system to predict flash floods one hour in advance. Geon-Wook Hwang, a researcher at the institute, explains: “A forecast, even if accurate, has no information value if it does not arrive early enough to significantly reduce the number of casualties and damage. damage caused by flash floods. »

You can follow THE MATERIAL In Facebook, X And Instagramor sign up here to receive our weekly newsletter.

_