AI weather forecasting could help to track extreme weather events such as the Los Angeles wildfires more accurately. Getty Images
AI weather forecasting could help to track extreme weather events such as the Los Angeles wildfires more accurately. Getty Images

'Nowcasting': how AI is reshaping weather forecasts



It is no secret that artificial intelligence is everywhere today. But, now, an AI front is passing over the world of weather forecasting.

AI has improved forecasting so much that, in some cases, it can outperform conventional systems, an expert has told The National.

Marouane Temimi, associate professor at the Stevens Institute of Technology in the US, said the pattern has emerged over the past few years and could lead to hugely improved short-term forecasting that can pinpoint which areas of a city can expect rain.

Speaking on the sidelines of the International Rain Enhancement Forum in Abu Dhabi on Wednesday, Prof Temimi said these forecasts – called nowcasting – could also lead to better emergency responses. They can be used for any extreme weather event from hurricanes to wildfires to storms.

But he also said that AI systems might run out of data if there is less of a commitment to conventional physics-based forecasting models.

“AI-based models are doing well, close enough to the physics-based model – the traditional model that we have been using,” said Prof Temimi. “In some cases they even overperformed them. The European model that was developed by the European Centre for Medium-Range Weather Forecasts has shown that.”

Where AI systems may have a crucial role is this ultra short-term “nowcasting” that can track a storm over several hours.

“They tend to have better accuracy when it comes to the location and the timing of the magnitude of rainfall,” he said, with such a model developed at his university. “If you have a last-minute decision’ to take – like evacuate people or close roads – then you rely on the nowcasting instead of the forecasting.”

Many technology companies are developing AI weather forecast models. An example is Graphcast developed by Google’s DeepMind that identified the landfall location of Hurricane Beryl last year before regular models.

Many countries are now exploring this development, with money pouring into the field and government agencies figuring out how to use such models. But much more research and funds are needed.

No data, no AI

But there is a much larger issue. Conventional forecasting models are a series of complex equations based on huge amounts of data built up over decades and compiled by experts.

“These are basically complex models that are built out of millions and millions of lines of codes and they try with their structure to mimic every single small process that happens out in nature … to predict weather,” said Prof Temimi, who is based in the New Jersey university's department of civil, environmental and ocean engineering.

He said there is an entire atmospheric science community that has been studying the physics of clouds, radiation of the Sun and much more that informs these physics-based models. This has been built up over decades thanks to this painstaking work.

But with the advent of AI over the past few years, things have started to change as it does not look at these microprocesses. It simply takes the information from satellite images to radar details and identifies patterns over time. “Unfortunately, AI does not go into those details,” he said. “AI can try to predict events without that strong knowledge of physics.”

The more data the better, but what happens when the data runs out? Prof Temimi said that data could become scarcer if there is a greater reliance on AI and less of a commitment to the hard task of gathering information without which the AI systems could not work.

“Eventually, we may risk running out of data to feed the AI models. The community will [need] to find a solution or a compromise because we cannot drop the physics-based model, in my opinion, and we cannot all shift to the AI models.”

The forum, meanwhile, has drawn experts in weather modification and water security from around the world.

Discussion on Wednesday focused on new cloud-seeding materials and the role of drones and aircraft in weather modification.

Student and early-career scientists from local and international research institutions also presented studies into the field. The forum continues until Thursday.

Updated: January 29, 2025, 2:43 PM