The European Centre for Medium-Range Weather Forecasting has partnered with one of China’s most successful technology firms to integrate its forecasting model into its services.
Pangu-Weather, developed by Shenzhen-based internet and communications giant, Huawei, is reportedly much faster at predicting weather patterns than typical forecasting methods used today.
Pangu-Weather was chosen because many meteorologists are beginning to realise that AI could play a significant role in the field, said Tian Qi, chief AI scientist with Huawei, in an August interview with the South China Morning Post.
Tian explained that the model is highly accurate, yet does not require massive computing power to apply.
The latest version of Pangu-Weather, which has been trained on more than nearly four decades of global weather data, outperformed what meteorologists consider to be the most sophisticated forecasting system, the numerical weather prediction (NWP) method, according to a July research paper published by the peer-reviewed journal, Nature.
The model can reportedly determine weather metrics like wind speed, humidity, air temperature and sea level barometric pressure on an hour-to-hour basis up to a week in advance and at a rate 10 000 times faster than the NWP method.
The newly-released AI forecasting tool predicted the pathing of Typhoon Doksuri, also known as Super Typhoon Egay, as it wreaked havoc in the Philippines, Vietnam, Taiwan and China, resulting in more than 100 fatalities and an estimated $15 billion worth of damage in just under two weeks.
Pangu-Weather provided crucial information that was used by several weather bureaus tracking the storm’s path, according to Tian.
Doctor Florian Pappenberger, a director at the European Centre for Medium-Range Forecasting, described AI-enhanced models like Pangu-Weather as the “quiet revolution” in the industry.
Data-driven weather prediction based on machine learning has tremendous potential. In addition, AI-powered, real-time analysis would be a game-changer compared to the NWP method, according to a paper Pappenberger and his colleagues published in July.
While the results are impressive, there is still work to be done, warns Huawei researcher, Xie Lingxi.
In a post on a Chinese social media platform, Zhihu, Xie, who earned his PhD in machine learning from the University of Tsinghua, told readers the AI systems used at present may underestimate typhoon strength as the dataset Pangu-Weather was trained on had a relatively low proportion of extreme weather events.
The NWP models should still be taken into consideration when making predictions about extreme weather events, added Xie.