Recent advancements in artificial intelligence (AI) have led to significant developments in weather forecasting, with new models demonstrating the ability to produce accurate forecasts much more quickly than traditional methods. A notable innovation is the Aardvark Weather model, developed by Richard Turner and his colleagues at the University of Cambridge. This model claims to generate comprehensive weather forecasts in approximately one second on a standard desktop computer, a stark contrast to the hours or days required by conventional numerical weather prediction (NWP) systems that rely on powerful supercomputers.
Historically, weather forecasting has depended on complex physics-based models that require extensive computational resources. In recent years, there has been a push to integrate AI into this process. Prior efforts by Google and DeepMind have showcased AI's potential to streamline forecasting by reducing the computational power needed for specific tasks within traditional models. The European Centre for Medium-Range Weather Forecasts has also adopted AI tools to enhance its forecasting capabilities.
However, Aardvark Weather distinguishes itself by replacing both the initialisation and forecasting stages of traditional models. It utilizes only 10 percent of the input data required by existing systems while achieving results comparable to the latest NWP forecasts. Despite its speed and efficiency, some experts caution that Aardvark's grid model, which is coarser than that of more established systems, may not capture complex weather patterns effectively.
Turner acknowledges that while Aardvark can outperform some existing models in identifying unusual weather phenomena, it still relies on physics-based models for training. As the field evolves, there is speculation that future advancements may enable AI to create independent forecasts that surpass current NWP methods, with a focus on innovative data structuring and neural network design.