Microsoft’s AI Model Is Outperforming Traditional Climate Forecasts

Climate-related disasters are increasing in frequency and severity, as evidenced by the floods in Valencia, the air quality crisis in New Delhi, and Hurricanes Helene and Milton in the eastern United States.

Natural disasters can lead to significant loss and damage to properties, infrastructure, and communities. Earth system forecasts play a vital role in preventing these substantial losses by offering early warnings for extreme events. These forecasts are essential for various critical applications, including agriculture, healthcare, and business.

However, traditional forecasting systems, although powerful, rely on highly complex numerical models developed over decades, requiring powerful supercomputers and large teams of experts.

An international team of researchers introduced Aurora, a state-of-the-art AI model designed to deliver faster, more accurate, and more affordable forecasts for air quality, ocean waves, extreme weather events, hurricanes, typhoons, and other related phenomena.

In a new paper published in Nature, the researchers explain how Aurora, as a large-scale foundation model, utilises the latest advancements in artificial intelligence to accurately predict various types of environmental events and potentially other Earth systems.

According to the study, Aurora was trained using over one million hours of diverse Earth system data and then fine-tuned for specific tasks. As a result, this AI model has, for the first time, outperformed several existing forecasting systems while being much faster and more efficient.

Here are examples of where Aurora excelled in critical forecasting domains:

• Air pollution: It provides 5-day global forecasts at a detailed scale, outperforming traditional, high-cost models for 74% of targets.
• Ocean waves: It gives 10-day global forecasts with better accuracy than existing models for 86% of targets.
• Tropical cyclones: It predicts storm paths more accurately than seven major forecasting centres in all cases tested.
• Global weather: It provides 10-day forecasts with excellent resolution, outperforming current best models on 92% of targets, including during extreme weather events.

The study highlights that Aurora is crucial for making accurate and efficient environmental predictions more widely available. This demonstrates how AI can transform climate and weather forecasting, providing high-quality information that is more accessible to a broader audience.

Aurora’s ability to outperform other operational forecasts at significantly lower computational costs represents a significant step toward democratising fast and accurate Earth system predictions. This advancement paves the way for broader access to reliable climate and weather information, which can significantly benefit developing countries that may lack expensive infrastructure and effective early warning systems.

Additionally, the researchers note that while they have only provided four applications of Aurora, it can be fine-tuned for a broader range of Earth system prediction tasks that may outperform current operational systems at significantly lower costs.

Potential applications of Aurora include predicting ocean circulation, local and regional weather, seasonal climates, vegetation growth and phenology, extreme weather events such as floods and wildfires, pollination patterns, agricultural productivity, renewable energy production, and sea ice extent.

Read the study to learn more about Aurora. Refer to the links in the sources provided below.