Environment Canada is moving to an AI predictive weather model. Here’s what that means

Research scientist Syed Husain explains how new AI-based models that learn from historical data will be used to forecast the weather.

A new artificial intelligence-driven weather model by Environment and Climate Change Canada could give Canadians earlier warnings for major weather events, which could potentially save lives, a scientist with the department says.

“The backbone of operational weather forecasting has been physics-based models … and solving complex mathematical equations,” Syed Husain said in an interview with CTV Your Morning Monday.

“Now … we have these AI-based weather models. Instead of learning from the laws of physics, they basically learn from historical weather data, and they analyze this data, detect patterns and then learn how to predict the future.”

Environment and Climate Change Canada announced plans in April to launch a new hybrid forecasting model that uses AI to analyze decades of historical weather data in minutes, identifying relationships between temperature, wind and pressure to better estimate future conditions, particularly for large-scale events like heat waves and hurricanes.

• READ MORE: Here’s how Environment Canada plans to use AI for its weather forecasting model

Husain said early research shows AI models have clear advantages.

“They are quite accurate … and much faster compared to physics-based models,” he said, but warned that AI on its own has limits, particularly when it comes to capturing the fine details needed to predict severe weather at a local level.

To address that gap, the department developed what Husain describes as a “best of both worlds” approach, meaning because the system is a hybrid, it also incorporates traditional physics-based modelling, which preserves small-scale details that are critical for forecasting extreme weather, such as strong winds or localized storms.

Husain said Canadians are unlikely to notice a difference in how forecasts look, but should benefit from improved accuracy.

“This model is going to predict with better accuracy, especially at medium- to long-range forecast,” he said. “This new hybrid model will provide them with alerts and warnings that are more timely, because our model can have a predictability gain of about half a day to one day.”

That added lead time could make a significant difference in a country as large and weather-prone as Canada, Husain said.

“Canada is a huge country with very active weather, so accurate and reliable weather forecasting is very important. For example, if we can have an alert for a major weather event coming, like half a day to one day earlier, that can save a lot of money, and that can save people’s lives.”

The department said the new system could make its six-day forecast as accurate as its current five-day outlook, a notable improvement in a field where gains have traditionally taken years to achieve.

Husain said the use of AI has dramatically sped up that progress.

“The kind of improvements that usually takes about seven to 10 years to achieve, with the help of AI, we were able to leverage that kind of accuracy gain over a period of about one to one and a half years.”