Artificial Intelligence (AI) has many use cases, including improved and efficient weather forecasting, which can be particularly useful for places experiencing regular typhoons like the Philippines.
In a recent development, Microsoft released its new AI model to beat the current forecasting methods in terms of air quality tracking, ocean waves, sea ice, weather pattern recognition, and tropical storms.
The new AI system, Aurora, is a large-scale model trained on more than one million hours of diverse geophysical data. It can produce forecasts for “any collection of Earth system variables at any desired resolution.”
According to the research journal Nature, Aurora achieved state-of-the-art performance in several forecasting domains, including a 5-day global air pollution forecast, outperforming existing numerical atmospheric chemistry simulations on 74% of targets.
It also generated a 10-day global ocean wave forecast at 0.25° resolution, a five-day tropical cyclone track forecast on 100% of targets, and a 10-day global weather forecast at 0.1° resolution, surpassing the modern numerical models on 92% of targets.
Aurora proved accurate, faster, and more cost-effective than the current technology. The computational costs alone of running Aurora are said to be a hundred times lower than those of the current system.
In one of its case studies, Aurora accurately predicted the landfall of 2023 Typhoon Doksuri (Bagyong Egay) in the Philippines with a 4-day lead time. The typhoon killed 137 people and caused over $28.6 billion in damages.
Compared to the track verification report from the National Hurricane Center (NHC) for the North Atlantic and East Pacific, Aurora outperformed all the headline models. The experiment follows the pioneering models of Huawei’s Pangu-Weather AI model in 2023.
While still in its earlier stages, this improvement in weather forecasting is a significant step forward towards natural disaster mitigation in the Philippines, particularly during the typhoon season, that are not only costly, but also fatal.