Do I need a brolly? Google uses AI to try to improve two-hour rain forecasts – The Guardian

Weather forecasts are notoriously bad at predicting the chances of impending rain – as anyone who has been drenched after leaving the house without an umbrella can testify.

Now, scientists at Google DeepMind have developed an artificial intelligence-based forecasting system which they claim can more accurately predict the likelihood of rain within the next two hours than existing systems.

Today’s weather forecasts are largely driven by powerful numerical weather prediction (NWP) systems, which use equations that describe the movement of fluids in the atmosphere to predict the likelihood of rain and other types of weather.

“These models are really amazing from six hours up to about two weeks in terms of weather prediction, but there is area – especially around zero to two hours – in which the models perform particularly poorly,” said Suman Ravuri, a staff research scientist at DeepMind in London and co-lead of the project.

“Precipitation nowcasting” is an attempt to fill this blind spot. Dr Peter Dueben, coordinator of machine learning and AI activities at the European Centre for Medium-Range Weather Forecasts, who was not involved in the research, said: “In nowcasting, what we try to do is to take observations from now, and try to make predictions of how the weather is going to look in a couple of minutes to a couple of hours. Machine learning can help you to build a tool that is extremely fast.”

DeepMind was not the only group that was attempting to develop such tools, but it was currently leading the field, he added. Its technology draws on high-resolution radar data, which can track the amount of moisture in the air by repeatedly firing a beam into the lower atmosphere and measuring the relative speed of the signal, which is slowed by water vapour.

Drawing on conversations with Met Office meteorologists about the types of weather prediction tools that would be most useful , Ravuri and his colleagues used a machine learning approach called generative modelling to develop a tool that could make probabilistic predictions of medium to heavy rainfall for the next 90 minutes, based on the past 20 minutes of high-resolution radar data.

As well as affecting individuals, heavy rain can disrupt transport and energy supply networks and agriculture.

DeepMind’s tool was evaluated alongside two existing rain prediction tools by more than 50 Met Office meteorologists, who ranked it first for accuracy and usefulness in 88% of cases. The results are published in Nature.

The DeepMind senior staff scientist Shakir Mohamed said: “AI has the potential to aid us in answering some of the most complex scientific questions in environmental science, such as climate change.

“This trial shows that AI could be a powerful tool right now by enabling forecasters to spend less time trawling through ever growing piles of prediction data and instead better understand the implications of their forecasts.”

Niall Robinson, the head of partnerships and product innovation at the Met Office, said: “Extreme weather has catastrophic consequences including loss of life and, as the effects of climate change suggest, these types of events are set to become more common. As such, better short-term weather forecasts can help people stay safe and thrive. This research demonstrates the potential AI may offer as a powerful tool for improving our short-term forecasts and our understanding of how our weather patterns are evolving.”

Dueben added that it was encouraging to see a big tech company such as Google working with expert meteorologists to develop new forecasting tools: “You can build the perfect tool, but if it is not going to be used by the forecasters it is pointless.

“I think this combination of the collaboration between Google and the Met Office, the involvement of the forecasters, and the new generative modelling approach which provides a new way to represent the distinct weather situations and the certainty of those predictions, makes this a significant step forward.”


Leave a Reply

This website uses cookies. By continuing to use this site, you accept our use of cookies.