AI Breakthroughs Could Improve Weather Forecasts

Your next weather update may be coming to you courtesy of artificial intelligence (AI). 

Britain's national weather service has developed an AI tool that it claims can accurately predict the likelihood of rain in the next 90 minutes. Making accurate weather predictions is a challenging problem that has resisted millenniums of effort. But researchers hope AI could revolutionize weather forecasting. 

"Any industry that is weather-sensitive is looking into ways to use AI to improve safety and operations," Renny Vandewege, the vice president of weather operations at data analytics company DTN, told Lifewire in an email interview. "For example, utilities are using AI to identify and predict grid resiliency and potential outages."

Nowcasting Rain

London is known for gloomy skies, but at least you may have a better warning when sprinkles start. Working with the UK's national weather service, AI company DeepMind has developed a deep-learning tool called DGMR for forecasting.

Experts judged DGMR's forecasts to be the best across a range of factors—including its predictions of the location, extent, movement, and intensity of the rain—89% of the time, according to a paper recently published in the journal Nature. The company calls the technique "nowcasting" because it's so timely. 

"We use an approach known as generative modeling to make detailed and plausible predictions of future radar based on past radar," DeepMind wrote on its website. "Conceptually, this is a problem of generating radar movies. With such methods, we can both accurately capture large-scale events, while also generating many alternative rain scenarios (known as ensemble predictions), allowing rainfall uncertainty to be explored."

Appu Shaji, an AI scientist not involved in the DeepMind study, called the company's work "impressive" in an email interview with Lifewire. 

"That being said, these works are still in their infancy, and we should expect to see considerable advancement in accuracy and forecasting possibilities in the coming years," he added. 

Predicting Chaos

Weather is a chaotic process that's difficult to predict with precision. 

"Advanced weather models and technology, like AI, improve forecasting to help us to better plan, prepare and lessen the impact of weather events," Vandewege said.

"As weather events become more frequent and extreme, accurate forecasts with a longer lead time mean businesses, communities and the public have more time and more information to make better decisions."

Weather simulations are currently run using computer models, Vikram Saletore, an AI expert at Intel, told Lifewire in an email interview. But, he said, weather models need to be run frequently as the environment changes for accurate forecasting. 

"AI dramatically improves weather forecasting by enabling and accelerating significantly these simulation environments to take in massive amounts of historical models with the current environment as input and run predictions on potential outcomes," Saletore added. 

Spire Global is one company that's already using AI programs to enhance forecasts. The PredictWind program provides wind forecasts to maritime and leisure sporting users by processing satellite data with computer algorithms. 

"Climate change is increasing the likelihood of extreme weather and global operations open businesses to the threat of weather disruptions anywhere in the world," Matthew Lennie, an AI expert Spire Global, told Lifewire in an email interview. 

Computing power has been a bottleneck for weather forecasting. As a result, some of the most powerful supercomputers have been built specifically to crunch forecasting numbers. 

"AI has an amazing chance to reduce this dependency on powerful engines and potentially run these models to get as good as or better results with significantly less computational load," Shaji said. "Deep learning doesn't try to solve these formulae directly, but predicts them based on observable patterns."

The AI method is similar to how stock market investors watch patterns over long periods, Shaji pointed out. "Deep learning has more accuracy," he added. "The predictive accuracy and capability of models will only become better in the future."