Weather forecasters rely on sophisticated models to make crucial decisions, often with life-or-death consequences. Any advancement in this field is welcome, and artificial intelligence (AI) holds significant promise to enhance forecasting accuracy and speed. In recent years, tech giants such as Google, Nvidia, and Huawei have introduced AI-based forecasting models. The latest innovation in this domain is Aurora, an AI weather model developed by Microsoft, which stands out for several key reasons, according to a report published in the journal Nature.
Aurora is already operational at one of Europe’s largest weather centers, where it runs alongside traditional and other AI-based models. What sets the Aurora model apart is its capability to deliver precise 10-day forecasts at smaller scales, a significant improvement over many existing models. Additionally, Aurora is not limited to weather predictions; it can be adapted to forecast any Earth system with available data. This versatility means it can predict phenomena like air pollution and wave height, in addition to weather events such as tropical cyclones. Users can expand its capabilities further, as one start-up has already tuned the model to predict renewable energy markets.
“I’m most excited to see the adoption of this model as a blueprint that can add more Earth systems to the prediction pipeline,” said Paris Perdikaris, a professor at the University of Pennsylvania who led Aurora’s development while at Microsoft. Aurora is also notably fast, delivering results in seconds compared to the hours that traditional models require.
Artificial Intelligence Gives Weather Forecasters a New Edge
Traditional weather forecasting models, utilized over the past 70 years, rely on complex mathematical equations to simulate the physical world—such as the sun heating the planet and winds circulating globally. Researchers input real-time weather data into these models to predict future conditions. Human forecasters then analyze results from multiple models, combining these with their expertise to provide the most likely weather scenario.
“Final forecasts are ultimately made by a human expert,” Dr. Perdikaris noted. This holds true even for AI-based forecasts.
The Limitations and Advantages of AI Models
AI weather forecasting models are quicker to develop, execute, and update. They are trained on vast amounts of weather and climate data to identify patterns, which are then used for predictions. However, AI models still require equation-based models and real-world data for initial conditions and validation.
“It doesn’t know the laws of physics, so it could make up something completely crazy,” said Amy McGovern, a computer scientist and meteorologist at the University of Oklahoma. Consequently, most AI weather forecasting models still depend on data from physics-based models, necessitating careful interpretation by human forecasters.
Dr. Perdikaris and his team constructed Aurora by training it on data from physics-based models before making purely AI predictions. They designed it to be applicable beyond weather, training it on diverse Earth system data sets to create a broad base of artificial expertise.
“Aurora is an important step toward more versatile forecasting systems,” said Sebastian Engelke, a professor of statistics at the University of Geneva. The model’s flexibility and resolution are its most unique contributions. While there has been significant momentum toward using AI for weather forecasting, major AI models are primarily global, not local. Forecasts for individual storms impacting cities still rely on specialized, traditional models.
The Future of AI in Weather Forecasting
Extreme weather events, such as heat waves or heavy downpours, remain challenging for both traditional and AI models to predict accurately. AI forecasting models require rigorous calibration and human verification before widespread adoption. Some models are already undergoing real-world testing. The European Center for Medium-Range Weather Forecasting, providing meteorological forecasts to numerous countries, has developed its own AI forecasting model, deployed in February. They now use this model alongside Aurora and other AI models for their weather services, with promising results.
“It’s absolutely an exciting time,” said Peter Düben, who leads the European center’s Earth modeling team. Despite some researchers’ cautious approach, given the necessary checks and improvements for AI models, the potential benefits outweigh the energy costs associated with training these tools, according to Dr. Perdikaris.
However, recent cuts to federal agencies, including the National Oceanic and Atmospheric Administration and the National Weather Service, could hinder further advancements in AI forecasting tools, as these agencies provide critical data sets and models necessary for development.
“It’s quite unfortunate, because I think it’s going to slow down progress,” Dr. Perdikaris commented.
Note: This article is inspired by content from . It has been rephrased for originality. Images are credited to the original source.