Environment and Climate Change Canada has launched a groundbreaking hybrid weather forecasting system that combines the power of artificial intelligence (AI) with the strengths of traditional forecasting methods. Through this strategic use of AI, Canada is enhancing public safety, improving emergency readiness, and giving Canadians more time to prepare for high-impact weather.
This advancement is the first of its kind in the world and demonstrates Canada's continued global leadership in weather and environmental prediction. With continued positive results in all stages of testing, Environment and Climate Change Canada's hybrid weather forecasting system is now operational.
Traditional weather forecasting is based on numerical weather prediction models. These models combine data from weather observations such as satellite data, weather balloons, and radars from domestic and other sources. Using this information, we can then simulate how the atmosphere will change over time, including temperature, wind, moisture, and precipitation.
Understanding our new hybrid approach
Environment and Climate Change Canada has developed a Canadian AI forecast model, the Global Environmental eMuLator (GEML). Building on the open source GraphCast model published by Google DeepMind, our scientists have developed innovative approaches to train and fine-tune GEML using historical weather data. Environment and Climate Change Canada research has shown that AI models such as GEML can predict large-scale weather patterns. These models can outperform traditional computer models with higher accuracy for these synoptic patterns, which are especially significant for longer lead times. However, a key limitation of the current generation of AI models is that they forecast only a limited number of variables, making it not suitable for operational needs. In addition, finer atmospheric details are smoothed out, which can reduce accuracy for certain types of extreme weather.
Building on this insight, our scientists conceived of a hybrid system, known as "spectral nudging". The hybrid system combines the strengths of AI models with our traditional models to improve predictability and accuracy, allowing us to leverage each approach's strengths. In doing so, it generates the complete set of variables which operational meteorologists can rely on daily.
In this hybrid approach, the traditional physics-based model is gently guided toward the atmospheric state of the GEML solution during the calculation. At the same time, the smaller-scale patterns that are important for heavy rainfall or severe storms are allowed to evolve based on existing, well-established mathematical and physical formulations.
The new High Performance Computing solution, operated by Shared Services Canada, combines enhanced computing capacity with AI capabilities to better support operations and advanced research and development. For more than a year, the hybrid system ran for extensive testing and validation. Forecasts from the hybrid system were systematically compared with those from the current operational system to ensure accuracy. Throughout this process, meteorologists closely monitored performance to ensure it produced reliable information for temperatures, precipitation, and storms across Canada, at both large and small scales.
Looking ahead
Canadians expect and deserve timely, accurate, and trusted weather forecasts year-round. As Canada continues to be a global leader in weather and environmental predictions, AI is expected to play an important role in improving these capabilities. This new hybrid approach enables us to optimize the use of emerging technologies, while leveraging the decades of scientific knowledge gathered in atmospheric and environmental science.
These advancements are not only important for day-to-day life and public safety, but also for effective planning in industries such as agriculture, transportation, construction, and energy. With the new hybrid system now in place, weather forecasters are bringing the benefits of this innovation directly to all Canadians.
