LONDON: When floods ripped through parts of Europe last month, the scale of the destruction took people by surprise. The intense rains should not have come as a surprise because those had been predicted by sophisticated forecasting systems enhanced with artificial intelligence.

But forewarned did not mean forearmed. Though the rains were accurately predicted, the effects in the deluged areas were not — a fact that highlights the difficulties of dealing with ever more common extreme weather.

Artificial intelligence has supercharged weather forecasting, using a range of statistical tools to analyse years of historical data and predict patterns, and at a lower cost than traditional numerical weather predictions.

For example, Google-funded GraphCast, a machine learning-based method trained dir­e­ctly from reanalysis data, was found to outperform traditional models. Reanalysis data relies on past forecasts rerun with modern forecasting models to provide the most complete picture of past weather and climate.

Experts stress need for better climate risk preparedness

But there are still gaps in knowledge, in how the information in used and in investment to strengthen data gathering models, experts say.

“In some cases and for some variables, AI models can beat physics-based models, but in other cases vice versa,” said Andrew Charlton-Perez, professor of meteorology at the University of Reading, in Britain.

One issue is that the effectiveness of an AI model is only as good as the information it is fed. If there is little input data, or extreme events happen more frequently at different times of the year or in different regions, weather disasters become more challenging to predict.

Communication is key

Since January, the European Centre for Medium-Range Weather Forecasts (ECMWF), an independent organisation that provides predictions four times per day to European countries, has been using the Artificial Intelligence-Integrated Fore­casting System (AIFS).

Click to see more

This data-driven forecasting model makes multiple predictions rapidly and delivers long-term forecasts of weather events like cyclones and heat waves.

The ECMWF readings before last month’s floods were accurate, experts say.

Thomas Wostal, press officer for meteorological observatory GeoSphere Austria, told Context-the Thomson Reuters Foundation that their numerical models, including the ECMWF’s predictions, foresaw 300-400 millimetres (11.8-15.7 inches) of rain locally, which came to pass.

But even with accurate forecasts, scientists say communication is the key, especially in an era when climate change means extreme weather is becoming more frequent.

“I think what happened with (the recent floods) … is that it’s so rare — a one in 150 to 200 a year event — that even if the weather models capture it, there’s a reasonable degree of uncertainty,” said Shruti Nath, a post-doctoral research assistant in predicting weather and climate at Oxford University.

Europe behind the curve?

Europe faces urgent climate risks that are outpacing policies and adaptation actions, a report from the European Environment Agency (EEA) has warned.

Extreme heat, drought, wildfires and flooding will worsen in Europe even under optimistic global warming scenarios and affect living conditions throughout the continent, the EEA says.

Some tech entrepreneurs say Europe is not ready.

Click to see more

Jonas Torland, a co-founder of Norway-based 7Analytics, which develops models for predicting floods and landslides, said governments and businesses in the United States had risk managers who were more accustomed to assessing environmental hazards, while in Europe, authorities lacked readiness.

“While AI is a crucial component of these models, unfortunately, governments are not investing in or purchasing these advanced AI solutions,” he said, adding that he believed governments “stick to their old data providers and consultants”.

Data processing is also a challenge because these complex AI models need to run updates every hour as forecasts change.

High computing power also means huge amounts of energy and water are needed, which makes AI models part of the problem because they are adding to the planet-heating emissions driving the climate emergency.

Some big technology companies, like Microsoft and Google, are exploring the use of nuclear power to run their huge data storage centres.

Other scientists stress that beyond refining their forecasting abilities, authorities need to invest in physical solutions, like developing areas where floodwater can safely be stored, and early warning systems.

They also need to minimise development in flood-prone areas, given the likelihood of more intense climate change-driven floods, and meet their commitments to limit emissions.

“It’s not a question of data or technology or knowledge. It’s a question of implementation, political will,” Friederike Otto, a senior lecturer at Imperial College in London, said in an email response to questions.

Published in Dawn, October 16th, 2024