SCIENCE GETS AHEAD OF UGLY WEATHER
BY JEANINE MOYER • PHOTO: PIXABAY
On the farm, decision-making is shaped by weather. Should the sprayer roll today or wait until a brewing storm passes? Can seeding be completed before incoming rain hits? Greater confidence in weather forecasting and earlier notice of developments could benefit farmers significantly.
A new Environment and Climate Change Canada (ECCC) initiative may give farmers just such an additional edge. The goal is to give meteorologists better information and, ultimately, to instill Canadians with greater confidence in the forecasts they rely on. On May 26, the department launched what it described as the world’s first operational hybrid weather forecasting system. It integrates the AI-based Global Environmental eMuLator to improve weather prediction accuracy. Rather than replace conventional forecasting, the approach combines AI with ECCC’s conventional physics-based weather modelling system to enhance forecast accuracy, particularly for medium- to long-range forecasts.
Employed for decades, these systems use complex mathematical equations to simulate unfolding weather patterns. Unlike AI-based forecasting, these digital platforms calculate the evolution of atmospheric conditions without the use of machine learning. This process requires an accurate picture of current conditions created with the use of data collected by aircraft, radar facilities, satellites and weather stations.
John Classen is chief meteorologist with CTV Edmonton. He pointed out these observations aren’t always immediately available. Additionally, to produce a forecast solely with the use of the older system requires enormous computing power and several hours of processing time, and the results aren’t always accurate. “As forecasts extend further into the future, the confidence in what the weather will look like is substantially reduced,” said Classen.
TWO MODELS, ONE FORECAST
“The AI modelling system is trained solely on data,” said Leo Separovic. An ECCC research scientist who helped develop the new system, he believes AI can strengthen prediction. Rather than calculate the complex equations that govern atmospheric physics, the AI system learns from patterns as it reanalyzes historical weather observations and forecast data collected over decades. By analyzing how weather has evolved in the past, the AI system learns to recognize patterns.
Conventional physics continues to do the heavy lifting in the new AI system. It is used to predict weather system development. While it provides accurate local detail, the AI model adjusts the large-scale picture. The hybrid system produces global forecasts with substantially improved accuracy, particularly in the five- to seven-day range. This delivers earlier prediction of weather systems. “It combines the strengths of both approaches,” said Separovic.
“The AI system has significant advantages, but it also has limitations. That’s why we use a hybrid model.”
While AI excels at pattern recognition and reduction of forecast errors, it cannot yet interpret every aspect of atmospheric development. The hybrid system still relies on non-AI modelling to distinguish between precipitation types such as rain, snow and hail, said Separovic. It also more accurately predicts high-impact weather events such as thunderstorms, although researchers continue to improve these capabilities. “It’s not perfect, but it’s improving the predictability of our weather,” he added.
Meaningful improvements to physics-based forecasting systems generally require years of development to improve the model itself and to better incorporate new weather observations. According to Separovic, the accuracy gains delivered by the development process of the hybrid system in less than two years is equivalent to a full decade of research and development with the previous system.
To train the ECCC’s hybrid model, researchers used a historical reanalysis dataset compiled by the European Centre for Medium-Range Weather Forecasts (ECMWF) and ECCC forecast data collected between 2019 and 2023.
“Because weather patterns evolve, and climate itself continues changing, the AI won’t simply learn once and stop,” said Separovic. “The model will need periodic retraining using newer Canadian data to maintain its performance.”
Although training AI requires substantial computing resources, the financial cost of the hybrid system is not significantly greater than the old system, said Separovic. The Canadian government is not alone in exploration of AI for weather prediction. Corporations such as Google and Nvidia, as well as public weather services such as ECMWF have spent years developing AI forecasting systems that partially informed Canada’s approach. Many national weather agencies are experimenting with hybrid approaches, said Separovic, but Canada is the first to operate a hybrid system. “Integrating AI into forecasting is a huge step and a significant improvement for forecasting here in Canada,” he said. “We’re extremely proud of this achievement.”
Because weather patterns evolve, and climate itself continues changing, the AI won’t simply learn once and stop. The model will need periodic retraining using newer Canadian data to maintain its performance.
—Leo Separovic
GREATER PREDICTABILITY
The hybrid model has delivered its greatest improvements in five- to seven-day forecasts, said Separovic. “The new system extends the useful predictability of the weather forecast by roughly one day compared to conventional physics-based forecasting alone,” he said. Forecasters can more confidently rely on the new hybrid system to identify severe weather patterns sooner. For farmers deciding when to spray, seed or harvest ahead of heavy rain or strong winds, an extra day could influence logistics and profitability.
Forecast accuracy in the 24- to 48-hour range, however, has changed little. Conventional forecasting performs well over short periods, noted Separovic. “But there is room for improvement with the hybrid approach.” Researchers now focus on the development of AI models that will improve these near-term forecasts.
While farmers may not immediately notice dramatic differences in the forecasts they receive, meteorologists almost certainly will. Classen said he hasn’t used the hybrid model in his daily forecasting yet. He does expect it will become a valuable tool that will improve accuracy and predictability and deliver earlier warning for major weather events. “It’s our job as meteorologists to read and interpret the information we receive from various models,” said Classen. To produce a forecast requires comparison of multiple weather models as well as experience and local knowledge. “That’s why every forecast is different.”
The hybrid system doesn’t disrupt weather prediction but is a natural evolution, said Classen. “This new system is 100 per cent a good thing. When you blend the two models, you get environmental science with speed.” He expects AI will deliver timely information and reduce the potential for human error.
THE FUTURE OF FORECASTING
While AI is expected to enhance the speed and accuracy of weather forecasting, Classen does not see it replacing meteorologists. He views it as a valuable resource with which to make more informed decisions. “Any time there is new information, our forecasts will improve, but you still need humans to interpret the data.”
Separovic agrees. Weather forecasting continues to depend on skilled professionals who understand atmospheric science, recognize when models are wrong and can communicate risk to the public. Although AI demonstrates tremendous promise, researchers are not ready to abandon existing forecast methods. Separovic cautioned that while the hybrid system is in its early stages, scientists, forecasters and the public will monitor its performance and progress. He predicts hybrid weather forecasting will become increasingly common as the technology improves and larger, higher quality datasets become available.
FORECAST MEETS FIELD
The real test for farmers will be whether better forecasts improve decision-making. In a written statement from Alberta Agriculture and Irrigation, press secretary Callum Reid noted dependable weather information remains essential to agriculture. “Farmers’ livelihoods depend on reliable, accurate weather information. While it’s still too early to say what the long-term impacts of the changes will be, we will keep a close eye on the new approach to ensure farmers have access to the dependable decision-making tools they need.”
ECCC shares this measured approach. Artificial intelligence won’t replace decades of atmospheric science. Rather, it is another tool to help forecasters extract more accuracy from an already sophisticated system.
AI NEEDS RADAR, TOO
As the ECCC hybrid AI weather forecasting system was rolled out, its High Impact Weather Research section was disbanded. The number of staff let go has not been disclosed, but this included the team responsible for weather radar research and came just a couple of years after Canada completed implementation of its $180-million new national radar network. The federal government recently installed 33 Doppler radar sites in 2024. The network will continue to operate, but critics say this is inadequate.
Dave Sills is director of the Northern Tornadoes Project and deputy director of the Canadian Severe Storms Laboratory. He noted that before its elimination, the radar research group worked to improve the quality of radar data, addressed technical and interference issues, developed radar processing and detection techniques and worked to get more information from the network for severe weather forecasting.
The new radar sites have capabilities that are not yet being fully realized, said Sills. This includes the improvement of detection and nowcasting of flash flooding, hail, tornadoes and other hazards. “The data coming out of the radars right now is OK, but they can produce so much more,” said Sills.
ECCC has said the radar network will be maintained, and that scientific and technical expertise will remain available to support its operation. “Maintenance and research are two different things,” said Sills. Without a dedicated research team, he’s concerned Canada could lose ground in the improvement of radar performance and warning capabilities.
“Open and unbiased research is essential to helping the agriculture sector remain strong and continue to grow,” said Alberta Agriculture and Irrigation press secretary Callum Reid in a written statement. “Farmers are unlikely to see any immediate changes to the weather information available to them.” He noted the many existing provincial agricultural weather resources will remain available. “However, this is concerning as reducing support for radar research could slow improvements in forecast accuracy, precipitation estimates and warning lead times. Alberta will closely monitor the impacts of the federal decision and will continue to advocate for research to ensure weather forecasts provide the information needed by Alberta’s producers and protect the public.”
Sills believes radar and AI are complementary, rather than competing, technologies. AI forecasting depends on high-quality observations, he noted, and radar provides some of the most important real-time information about severe storms. “The two work together,” he said. “AI still needs good data.”
The question is not simply whether Canada’s new radar sites will continue to operate, said Sills. It is whether Canada will have the scientific expertise needed to continually improve the system and provide better weather information and earlier warnings.
Comments