BEIJING, June 30 (Xinhua) -- As the world remained lost in the embrace of slumber, 64-year-old fisherman Julian Bata and his wife awoke to the howling typhoon. The sound was all too familiar to the residents of the coastal community on Manaet Island, the Philippines. Due to its unique topography and location, the island faces multiple natural disaster risks.
Before evacuating with his family, Julian rushed to check on his fishing boat. Fishing is the primary source of income for these fishermen, who rely on the sea for their livelihood. "Weather forecasts are crucial," Julian said. "They determine whether or not we go out to sea and where we go. We must check them before every trip."
Weather forecasts serve as a compass and reassurance for fishermen, and even more so, as a lifeline. For years, Dr. Bai Lei, a young scientist at the Shanghai Artificial Intelligence Laboratory, has been closely following global climate change. The devastation caused by natural disasters has left him heartbroken, but it has also strengthened his resolve to pursue his AI weather forecasting model project.
AI Weather Forecast Ensuring a Safe Return for Those at Sea
The 2023 assessment report of the Intergovernmental Panel on Climate Change (IPCC) indicated that global warming is projected to reach 1.5 degrees Celsius between 2021 and 2040. Against this backdrop, the theme of World Meteorological Day 2024 is "At the Frontline of Climate Action," emphasizing the urgency of collaborative climate action. Timely dissemination of weather forecasts and warnings has become crucial in addressing global climate change and its impacts.
The western North Pacific region adjacent to China experiences more than one-third of the world's annual typhoons, making accurate typhoon forecasting critical for mitigating disaster losses. The advent of the Fengwu weather model coincided with the landfall of Typhoon Doksuri, the fifth typhoon of the season. Fengwu reduced the error margin to 38.7 kilometers 24 hours in advance, surpassing the accuracy of 54.11 kilometers achieved by the European Centre for Medium-Range Weather Forecasts, a leading international weather forecasting authority, and 54.98 kilometers achieved by the U.S. National Centers for Environmental Prediction. "Fengwu weather model stood out in comparative tests," remarked Melinda Peng, a professor at the University of Colorado, expressing her hope to collaborate with the team on exploring weather models.
The World Meteorological Organization has said that over the past few decades, the effectiveness of global medium-range weather forecasts has improved by only one day every decade, due to the accuracy level of meteorological observations, the complexity of physical processes in atmospheric systems, and the immense computational resources required to solve atmospheric models. In contrast, the AI-powered weather model Fengwu achieves high-precision global weather forecasts for the next ten days in just 30 seconds. "Fengwu significantly outperforms traditional models in terms of efficiency," said Bai.
In August 2023, Florian Pappenberger, deputy director-general of the European Centre for Medium-Range Weather Forecasts, participates in the AI for Earth forum. (Provided by Shanghai Artificial Intelligence Laboratory)
In 2022, Bai established the Fengwu weather and climate large model research team. The name "Fengwu" is derived from the "Xiangfeng Tongwu" of the Qin and Han dynasties, meaning a crow-shaped bronze wind detector. It was the world's earliest wind measuring device. The weather forecast large model Fengwu not only embodies the wisdom of ancient Chinese people but also symbolizes the laboratory's dedication to breakthroughs and relentless exploration in the field of AI for Science, with meteorology as a representative.
"The development of AI-assisted weather forecasting" was named one of the top ten scientific breakthroughs of 2023 by the American journal Science. Large models will play an important role in this field. To achieve a truly user-friendly experience for a global weather forecast large model, it requires continuous innovation and meticulous attention to detail from the research team.
In 2024, the Shanghai Artificial Intelligence Laboratory collaborated with domestic and international research institutions to release the global high-resolution AI weather forecast large model "Fengwu GHR." For the first time, AI was used to achieve 10-kilometer modeling and forecasting of medium-term weather. The research team used an original AI algorithm to improve the forecasting resolution of this large model to 0.09 degrees of latitude and longitude (9 km by 9 km), corresponding to a ground surface area of around 81 square kilometers, which is seven times larger than the accurate forecasting range of the first-generation "Fengwu."
"Today, the fully upgraded 'Fengwu GHR' is equivalent to a 4K high-definition large model, which will help achieve more refined and efficient weather forecasting," said Dr. Ouyang Wanli, a leading scientist at the Shanghai Artificial Intelligence Laboratory.
Dr. Bai Lei discusses the development of AI-assisted weather forecasting with guests at the AI for Earth forum of the 2023 World Artificial Intelligence Conference. (Provided by Shanghai Artificial Intelligence Laboratory)
"The 'Fengwu GHR' has gradually developed into a weather and ocean forecast system composed of multiple models, including short-term, medium-term, and multi-year atmospheric and ocean comprehensive forecast systems," said Bai when introducing the large model at the 2024 Global Developer Conference.
At the 2024 Global Developer Conference, Bai Lei introduces the weather and ocean forecast large model. (Provided by Shanghai Artificial Intelligence Laboratory)
Exploring the Boundaries of AI Weather Forecasting Capabilities
Professor Niklas Boers from the Technical University of Munich introduced that the climate forecast version of Fengwu, jointly developed by the two parties, has been completed. It uses AI methods to predict ocean climate. The prediction effect of this model exceeds that of existing ocean models and further improves human capabilities to predict sub-surface ocean heatwaves, El Niño and other climate events.
In August 2023, Niklas Boers of the Technical University of Munich presents the climate forecast version of Fengwu. (Provided by Shanghai Artificial Intelligence Laboratory)
In August 2023, during the annual meeting of the United Nations Economic and Social Commission for Asia and the Pacific/World Meteorological Organization (WMO) Typhoon Committee, the Fengwu large meteorological model team introduced the application of the large model, which attracted the attention of meteorological departments from countries such as Cambodia and Vietnam. At the same time, the Fengwu team reached a consensus with meteorological research institutions to jointly explore the application of Fengwu in daily forecasting and the forecasting of typhoons and other extreme weather conditions in Pakistan and Southeast Asian nations.
"Artificial intelligence methods can solve many challenges in the field of meteorology, and we have seen good proof through the Fengwu large meteorological model. We are looking forward to exploring more accurate and timely weather forecasting systems," said Florian Pappenberger, deputy director-general of the European Centre for Medium-Range Weather Forecasts.
The world is changing rapidly. Fengwu can apply AI methods to a wider range of research in earth science fields such as meteorology, environment, astronomy and geology. AI can also help with major needs such as the pursuit of "carbon neutrality", disaster reduction and energy security. With the advent of the AI era, the future is no longer a dream! (Luo Hao)
A single purchase
