Artificial intelligence improves the accuracy of short-term weather forecasts and especially the prediction of storms and heavy rains, according to research by the British Meteorological Office and DeepMind, the London-based AI company.
Their project focused on ‘now broadcast’ – the timing, location and intensity of high-resolution precipitation of up to two hours ahead – which is not well handled by the supercomputer models used to scale the weather on a larger scale during to predict the next day or week.
This is critical for applications, ranging from alerting emergency services to an impending flood risk, and letting outdoor event organizers know when a downpour is coming.
The results, published in Nature, shows that an AI-based approach called ‘deep generative modeling’ or DGM performed better than other new effects methods across a wide range of measures. It was ranked first for accuracy and usability by 89 percent of a panel of 56 professional meteorologists, blinded to the source of the forecasts.
“Improving the accuracy of short-term forecasts is an incredibly important endeavor,” said Niall Robinson, Head of Partnerships and Product Innovation at the Met Office, Britain’s national weather and climate service.
“Extreme weather has catastrophic consequences, including loss of life, and as the effects of climate change indicate, these types of events will become more common,” he added. “This research demonstrates the potential AI it can offer as a powerful tool to improve our short-term forecasts and our understanding of how we develop weather patterns.”
The DeepMind and Met Office researchers trained DGM to predict the evolution of precipitation (rain and snow) by analyzing British radar maps for three years. It shows how much rain falls every five minutes at a spatial resolution of 1 km.
According to Shakir Mohamed, a senior DeepMind researcher, the AI technique differs greatly from two other areas in which DeepMind has been successful: to play games as Go and determine the shape of protein molecules. The company was founded in 2010 and acquired by Google in 2014.
DGM focuses on the probability that rows play out like rain patterns rather than achieving a specific result, such as winning a match or discovering how a protein folds.
Robinson said the Met Office is considering how to use the DeepMind research in its operational forecasts. “We need to carefully consider how new tools are deployed and maintained, the best user interfaces for our meteorologists and how they fit into all the other forecasts we provide,” he said.
Looking ahead, Robinson said the Met Office “is currently investigating where we can do next work on other research questions. [with DeepMind], including the application of AI to determine the impact of climate change and the processing of weather observations to create even better weather forecasts. ”
The researchers said the goal was not to replace human meteorologists with automated experts, but to improve their work.
“AI can be a powerful tool that enables forecasters to spend less time through ever-growing stacks of forecast data and rather focus on a better understanding of the implications of their forecasts,” Mohamed said. ‘It will be an integral element to mitigate the adverse effects of climate change today, support adaptation to changing weather patterns and potentially save lives.