Microsoft has developed an artificial intelligence (AI) model called Aurora, which has been shown to outperform traditional forecasting methods in tracking air quality, weather patterns, and climate-related tropical storms. The study, published in the journal Nature, reveals that Aurora generated 10-day weather forecasts and predicted hurricane trajectories more accurately, faster, and at a lower cost than conventional forecasting methods.
Key Findings
According to the study’s senior author, Paris Perdikaris, an associate professor of mechanical engineering at the University of Pennsylvania, Aurora is the first AI system to consistently outperform all operational centers for hurricane forecasting. Trained solely on historical data, Aurora accurately forecasted all hurricanes in 2023 more precisely than operational forecasting centers, such as the U.S. National Hurricane Center.
Improved Accuracy and Efficiency
Aurora’s computational costs were several hundred times lower than traditional weather predicting models, which are based on physical principles like conservation of mass, momentum, and energy, and require significant computer power. The model’s performance follows the success of the Pangu-Weather AI model developed by Huawei in 2023, potentially marking a paradigm shift in how major weather agencies forecast weather and extreme events exacerbated by global warming.
Outperforming Traditional Models
Aurora outperformed the European Centre for Medium-Range Weather Forecasts (ECMWF) model in 92% of cases for 10-day global forecasts, on a scale of approximately 10 square kilometers. The ECMWF is considered the global benchmark for weather accuracy. In a notable example, Aurora correctly forecasted four days in advance where and when Typhoon Doksuri, the most costly typhoon ever recorded in the Pacific, would hit the Philippines, whereas official forecasts at the time had it heading north of Taiwan.
Future Implications
The promising performance of Aurora and similar AI models is being closely scrutinized by weather agencies. Several agencies, including Meteo-France, are developing their own AI learning models alongside traditional digital models. Florence Rabier, Director General of the ECMWF, noted that their first ‘learning model,’ made available to member states in February, is ‘about 1,000 times less expensive in terms of computing time than the traditional physical model.’
While AI models like Aurora show great potential in improving weather forecasting, they also raise questions about the future of weather prediction and the potential for AI to aid in humanitarian responses. As Perdikaris stated, ‘I believe that we’re at the beginning of a transformation age in air system science.’ The next step is to develop systems that can directly work with observations from remote sensing sources to generate high-resolution forecasts anywhere.