AI has transformed weather prediction — don't let the negativity win

AI has plenty of risks that we must guard against, but data-driven AI is a tool we should embrace to protect our infrastructure, our climate, and our society, writes Andrew Parnell 

Together, we are looking at how to use AI to predict extreme weather, flash floods, heavy rainfall, storms, and power cuts. Picture: Chani Anderson

Right now much of the conversation about Artificial Intelligence (AI) is focussed on the negatives. We hear a lot about data issues, hallucinations, or the large emissions associated with training the models. 

These are important worries, but they ignore a quiet revolution happening right now. Over the last three to four years, AI has completely transformed how we predict the weather. Far from being a threat, this technology could be one of our best tools for handling the climate and biodiversity crises on our island.

I lead the Aimsir research centre at University College Dublin, a project funded by Met Éireann that started in September 2025. We are running 20 projects with experts from UCD and Met Éireann. 

Together, we are looking at how to use AI to predict extreme weather, flash floods, heavy rainfall, storms, and power cuts. Our main goal is simple: Create the best possible AI tools to forecast our weather as accurately as possible.

To understand how massive this AI shift is, we have to look at how weather forecasting used to work. For the last 70 years, we relied on complex mathematical equations to map out the atmosphere. 

This physics-based method does an incredible job and saved countless lives around the world.

It’s amazing to consider how much our predictions have improved over the decades. In the 1960s if an Atlantic storm was heading our way, our estimate of where it would land three days later was off by more than 700 kilometres. 

By the early 2020s thanks to satellites and modern supercomputers, that three-day error dropped to just 100 kilometres. Today our short-term one-day forecast error is now under 50 kilometres.

AI weather forecasting

These achievements were built on the traditional physics approach. But we have now hit a point where AI can do the job even better. In 2022, the European Centre for Medium-Range Weather Forecasting released a new AI forecasting system called AIFS. 

Depending on how you measure it, this AI model beats our best traditional forecasting methods by 10% to 15%. It is not a futuristic concept; it is being used right now across Europe to help create the daily forecasts you see on your phone or television.

The most mind-blowing thing about this new AI method is how efficient it is. It can generate a global weather forecast 1,000 times faster than older physics models. 

Instead of waiting for hours, an AI forecast takes less than a second. In fact, it can run on a standard laptop that you or I might have on our desks.

Andrew Parnell: 'AI has plenty of risks that we must guard against, but data-driven AI is a tool we should embrace to protect our infrastructure, our climate, and our society.'

This massive jump in speed brings two huge benefits. First, because these AI models run so quickly, they use a fraction of the computer power. This means the carbon footprint of running the forecast is much smaller than the older methods. 

Second, systems like AIFS are open source, meaning anyone in the world can download the weather forecast model for free. For a long time only wealthy countries could afford the massive supercomputers needed for top-tier weather forecasting, but now anyone with a decent computer can train their own weather model.

To prove just how accessible this technology is, look at how the entire AI weather field started. In 2021, Ryan Keisler decided to experiment with weather data in his spare time. He used a basic gaming PC, applied a few new AI methods, and kickstarted a global scientific revolution.

It is important to understand that this type of technology is quite different from Large Language Models (LLMs) like ChatGPT. I like to think of these weather models as ‘data-driven’ AI (sometimes called Machine Learning), as opposed to language-driven AI like ChatGPT.

While they share some of the same underlying maths, data-driven AI is far more targeted. The datasets are much smaller. 

The energy needed to train the AI is tiny compared to ChatGPT, and it actually uses fewer resources than our traditional physics-based weather forecasts. Models like these are a clear benefit to society.

Ireland and AI

Ireland has been a leader in data-driven AI for decades, largely thanks to government-funded research centres and training programmes. 

If we want to keep this lead and make our weather forecasts even more accurate, we must make sure we keep funding this research area.

One of the most important aspects of this funding drive is the new CASPiR High-performance computing system acquired through the Irish Centre for High-End Computing. 

What's your view on this issue?

You can tell us here

This system is absolutely essential if Ireland is to stay at the cutting edge of AI prediction.

We have a fantastic education system that is turning out brilliant, highly-skilled graduates. As a country, we need to give them the backing they need. 

AI has plenty of risks that we must guard against, but data-driven AI is a tool we should embrace to protect our infrastructure, our climate, and our society.

• Andrew Parnell is the Professor of Data Science for Weather and Climate at University College Dublin and director of the Aimsir and Decarb-AI research centres. He is deputy director of the Climate+ Co-Centre in Climate, Biodiversity and Water.