A joint team from the University of Alberta and Environment Canada has deployed a neural network that processes 47 petabytes of historical atmospheric data to produce month-ahead forecasts that outperform existing models by a factor of three.
The system, dubbed "AuroraNet," could revolutionize agriculture, disaster preparedness, and energy planning across Canada. Unlike previous forecasting tools that relied exclusively on physical atmospheric simulations, AuroraNet uses a hybrid approach that blends physics-based models with deep pattern recognition trained across six decades of global weather data.
— Dr. Liwei Chen, Lead Researcher
The implications for Alberta's agricultural sector alone are enormous. Grain growers, oilseed farmers, and market gardeners could now make planting and harvesting decisions with a degree of certainty previously impossible. Early access partners in Lethbridge and Red Deer reported that the tool correctly predicted a late frost in March with 27 days of advance notice — in line with the system's stated accuracy.
Environment Canada plans to integrate AuroraNet into its public forecasting infrastructure by Q3 2026, with a consumer-facing app expected to follow in early 2027.