An artificial intelligence model recently deployed at the Hong Kong Observatory and mainland China’s National Meteorological Centre can solve one of the toughest challenges in weather forecasting: predicting when a typhoon will rapidly intensify.
Li Qinglan, a professor at the Shenzhen Institutes of Advanced Technology (SIAT) who is leading the project, said the system was installed around three weeks ago and had provided “real-time updates on the progression of Typhoon Jangmi”.
Jangmi, which formed late last month and made landfall in Japan on June 3, forced Hong Kong carriers including Cathay Pacific Airways and Hong Kong Airlines to cancel or reschedule flights to Japan.
The observatory has predicted that Hong Kong will experience four to seven typhoons between now and October, and has warned that some could become super typhoons due to the El Nino phenomenon.
Forecasting rapid intensification, when a tropical cyclone’s maximum sustained winds increase by 15 metres per second (49.2 feet per second) within a 24-hour period, or by 10m/s within 12 hours, has been one of the toughest challenges in meteorology.
“Rapid intensification rarely happens, and is highly unpredictable, making preventive measures and responses extremely likely to be delayed,” Li said in a statement issued by SIAT last week. SIAT is affiliated with the Chinese Academy of Sciences.
Traditional numerical weather prediction technology could not accurately reflect the evolution of typhoon intensity, she said, while the statistical-dynamic method failed to capture the non-linear characteristics of typhoon intensity changes.
The new model uses machine learning, a field of AI where algorithms learn the patterns of given data and make predictions about new ones. Li’s team, which started research on typhoon intensity in 2013, integrated four types of machine learning algorithms to improve forecast accuracy.
It used the model to simulate results for 24-hour rapid intensification in the North Atlantic Ocean between 2016 and 2020. Compared with forecasts from the US National Hurricane Centre (NHC), it “achieved a higher hit rate and a lower false alarm rate”, SIAT said.
Wong Wai-kin, a senior scientific officer working on forecast development at the observatory, said the team would continue to use real-time data to evaluate the model’s performance.
He said the observatory used a variety of AI tools to support weather forecasting, with some forecast products powered by AI models made available to the public on its Earth Weather web portal.
The United States has also been developing AI for weather forecasting. The NHC said on its website that it had teamed up with Google DeepMind to develop a new AI hurricane forecast model, which was used experimentally last year.
Chinese meteorologists have also used AI in other areas. Two years ago, the Shanghai AI Laboratory, Hong Kong University of Science and Technology and other organisations launched FengWu-GHR, an advanced, AI-driven global high-resolution weather forecasting model that extended the effective forecast lead time by 12 hours to 11.25 days.
Li said her team was also working on the use of AI for forecasting strong winds, heavy rain and storm surges. -- SOUTH CHINA MORNING POST
