Canada makes its AI-hybrid weather model operational to give people more warning of severe storms

Environment and Climate Change Canada said it has made operational a new hybrid weather-forecasting system that combines the power of artificial intelligence with the strengths of traditional, physics-based forecasting. After positive results across testing stages, the agency described the move as the first of its kind in the world, aimed at giving Canadians more time to prepare for high-impact weather. The development was widely reported in mid-May 2026.

At the heart of the system is a model the agency calls GEML, the Global Environmental eMuLator. Canadian scientists built it on GraphCast, the open-source AI weather model published by Google DeepMind, then trained and fine-tuned it on historical Canadian weather data. AI models like GEML are especially good at predicting large-scale, synoptic weather patterns and tend to outperform traditional computer models for those patterns, particularly at longer lead times when early warning matters most.

“After positive results across testing stages, the agency described the move as the first of its kind in the world, aimed at giving Canadians more time to prepare for high-impact weather.”

The "hybrid" design is the clever part. Rather than throwing out decades of work, the agency keeps its traditional GEM physics model running alongside GEML, because the physics model preserves the small-scale local details that pure AI models tend to smooth away. Combining the two lets forecasters predict extremes such as strong winds or heat waves more accurately than either approach alone, blending the AI's grasp of the big picture with physics-based fine detail.

There are sensible caveats. AI weather models inherit the biases of the historical data they learn from, they still rely on traditional observations and physics, and a better forecast only protects people if warnings reach them and they act. Forecasting also remains probabilistic, not a guarantee. But building on an openly published model rather than a closed system, and keeping trusted physics in the loop, is a thoughtful way for a public weather service to adopt AI, one squarely aimed at the public-safety goal of giving communities a little more time before the storm arrives.