Hong Kong’s Leap into AI-Powered Typhoon Forecasting
Hong Kong, a city long vulnerable to the destructive forces of typhoons and extreme weather, is embracing artificial intelligence (AI) to revolutionize its weather forecasting capabilities. The Hong Kong Observatory (HKO) has announced significant investments in AI technology, aiming to improve the accuracy and timeliness of warnings for tropical cyclones, rainstorms, and other severe weather events. This move comes after AI models demonstrated notable success during recent record-breaking rainfall and typhoon events, even as challenges remain in pinpointing the intensity and exact locations of impacts.
As climate change intensifies storms and increases the unpredictability of weather patterns, the stakes for accurate forecasting have never been higher. Hong Kong’s efforts are part of a broader regional and global shift toward AI-driven meteorology, with neighboring Taiwan and leading research institutions worldwide also leveraging machine learning to enhance disaster preparedness.
Why Is Accurate Typhoon Forecasting So Critical for Hong Kong?
Hong Kong’s subtropical location places it squarely in the path of Pacific typhoons, which can bring devastating winds, torrential rain, flooding, and landslides. The city’s dense population and complex infrastructure mean that even small improvements in forecast accuracy can save lives, reduce economic losses, and help authorities better coordinate emergency responses.
Recent years have seen a string of extreme weather events. In August, Hong Kong experienced its highest daily rainfall since records began in 1884, prompting the city’s second-longest black rainstorm warning. The aftermath included severe flooding, landslides, and widespread disruption. In neighboring regions, deadly floods in Gansu province and mass evacuations in Guangdong underscored the urgent need for better early warning systems.
Traditional forecasting methods, while effective, have limitations—especially when it comes to predicting the precise path and rapid intensification of tropical cyclones. This is where AI promises to make a difference.
How Does AI Improve Weather Forecasting?
AI-based weather forecasting relies on machine learning algorithms trained on vast amounts of historical weather data. Unlike traditional numerical weather prediction (NWP) models, which simulate atmospheric physics using complex equations and supercomputers, AI models learn patterns and relationships from decades of global weather observations. Once trained, these models can generate forecasts much faster and with less computational power.
For example, the current generation of AI models used by the HKO are trained on approximately 40 years of global reanalysis data—amounting to about 25 trillion data points. These models can analyze and predict the evolution of weather systems, including typhoons, by recognizing subtle cues that may elude conventional models.
According to Lo Ka-wai, a scientific officer at the Observatory, “By learning from this extensive dataset, the AI models can identify the patterns and relationships of weather evolution, which can then be used to predict future weather changes.”
AI models have already shown their value in operational settings. During the approach of Tropical Cyclone Wutip, for instance, AI-powered forecasts provided a more accurate prediction of the storm’s trajectory than traditional models, helping the Observatory issue timely warnings and avoid unnecessary disruptions.
AI Models in Action: The Case of Typhoon Wutip
In June, as Typhoon Wutip threatened Hong Kong, the Observatory relied on both traditional and AI-based models to forecast its path. While a European model predicted Wutip would approach the southern coast of China and an American model suggested a closer approach to Hong Kong, two AI models estimated a more westerly track—closer to what actually occurred. As a result, Hong Kong was spared a direct hit, and the Observatory’s warnings reflected the more accurate AI predictions.
This success was not an isolated incident. AI models have repeatedly matched or exceeded the accuracy of traditional models in predicting typhoon tracks, particularly for storms in the Western Pacific. The Observatory has since expanded its use of AI, integrating models such as Fuxi (developed by Fudan University), Pangu-Weather (by Huawei), and the Artificial Intelligence Forecasting System (AIFS) from the European Centre for Medium-Range Weather Forecasts (ECMWF).
What Are the Strengths and Limitations of AI Forecasting?
AI models offer several advantages over traditional methods:
- Speed: AI can generate forecasts in minutes, compared to hours for physics-based models.
- Energy Efficiency: AI models use far less computational power, making them more sustainable.
- Track Accuracy: AI models often outperform traditional models in predicting the path of tropical cyclones.
- Early Warning: AI can detect patterns and issue warnings earlier, giving authorities and the public more time to prepare.
However, challenges remain. AI models tend to underestimate the intensity of tropical cyclones, particularly during rapid intensification events—a phenomenon where a storm’s wind speeds increase dramatically in a short period. This is a critical limitation, as nearly 80% of major hurricanes and typhoons undergo rapid intensification, and the number of such events is rising due to climate change.
Another challenge is the limited ability of AI models to predict extreme weather events that are rare or unprecedented, as these events may not be well-represented in the historical data used for training. For example, during the record-breaking rainstorm in September 2023, AI models provided useful guidance but struggled to capture the full extent of the rainfall and its precise location.
As the Hong Kong Observatory notes, “The capability of AI models in predicting extreme weather events has yet to be verified because of their rare occurrence in history, not to mention record-breaking weather events were not present in AI models’ training data.”
Comparing AI Models: Global and Regional Perspectives
Recent scientific studies have compared the performance of leading AI weather models in East Asia and the Western Pacific. Five major models—Pangu-Weather, FourCastNet v2, GraphCast, FuXi, and FengWu—were evaluated for their ability to predict typhoon tracks and intensity. FengWu emerged as the most accurate for track predictions, while FuXi and GraphCast also performed well. However, all models struggled with intensity forecasts, highlighting the need for further research and higher-resolution regional models.
Ensemble approaches, which combine the outputs of multiple models, have shown promise in improving forecast reliability. The rapid progress in AI-driven meteorology suggests that continued collaboration between AI researchers and meteorological agencies will be key to overcoming current limitations.
How Is Hong Kong Integrating AI into Its Forecasting Operations?
The Hong Kong Observatory has been at the forefront of AI adoption in meteorology. Since mid-2023, the HKO has introduced several AI models into real-time trial operations, making their forecasts available to the public through the Earth Weather webpage and the MyObservatory app. The Observatory uses AI not only for typhoon tracking but also for predicting rainfall, wind, temperature, and humidity up to fifteen days in advance.
He Yuheng, acting senior scientific officer at the HKO, explained that AI models successfully predicted the development of a rainband that brought heavy rain to the Guangdong coast a week before the August downpour. While the models were not perfect in pinpointing the exact locations of the heaviest rain, they provided a valuable basis for issuing timely warnings and preparing emergency responses.
The Observatory is also collaborating with international partners and private companies to stay at the cutting edge of AI forecasting. For example, Google DeepMind’s Weather Lab and ECMWF’s AIFS are being tested alongside local models, offering new tools for forecasters to evaluate and compare.
Regional Adoption: Lessons from Taiwan and Beyond
Hong Kong is not alone in its AI journey. Taiwan’s Central Weather Administration has also adopted AI models to track typhoons, with impressive results. During Typhoon Gaemi, AI-based forecasts predicted a direct hit eight days in advance, outperforming conventional models and giving authorities more time to prepare. According to Lu Kuo-Chen, deputy head of the CWA, “AI boosted forecasters’ confidence and gave authorities more time to prepare.”
Private sector involvement is accelerating progress. Companies like Nvidia, Google, and Microsoft are developing advanced AI weather models and partnering with meteorological agencies worldwide. These collaborations are driving innovation and helping to address the challenges of intensity forecasting and local impact prediction.
Challenges and the Road Ahead
Despite the promise of AI, several hurdles remain before it can fully replace or surpass traditional forecasting methods:
- Data Quality and Availability: AI models depend on high-quality, comprehensive datasets. Gaps in historical data or inaccuracies can limit model performance.
- Intensity Forecasting: Predicting rapid intensification and peak wind speeds remains a major challenge for AI and traditional models alike.
- Local Impacts: High-resolution regional models are needed to predict localized effects such as flash floods and landslides.
- Human Oversight: Expert meteorologists are still essential for interpreting AI outputs, especially in high-stakes situations.
As the field evolves, the Hong Kong Observatory and its partners are committed to ongoing research, model development, and public education. The goal is not only to improve forecast accuracy but also to build public trust and ensure that warnings are understood and acted upon.
Broader Implications: AI and Climate Resilience
The integration of AI into weather forecasting is more than a technological upgrade—it is a vital step toward building climate resilience in an era of increasing extremes. As storms become more intense and unpredictable, early and accurate warnings can mean the difference between safety and disaster for millions of people.
Hong Kong’s experience offers valuable lessons for other cities and regions facing similar threats. The rapid adoption of AI, combined with international collaboration and public engagement, demonstrates a proactive approach to managing climate risks. As AI models continue to improve, they will play an increasingly central role in safeguarding communities and supporting sustainable development.
In Summary
- Hong Kong is investing in AI to improve typhoon and extreme weather forecasting, following successful trials during recent storms.
- AI models have outperformed traditional methods in predicting typhoon tracks but still face challenges in forecasting intensity and local impacts.
- The Hong Kong Observatory is integrating multiple AI models into its operations and collaborating with international partners.
- Regional neighbors like Taiwan are also leveraging AI, with private companies accelerating innovation in the field.
- Ongoing research and human expertise remain essential to address the limitations of AI and ensure effective early warning systems.
- The adoption of AI in meteorology is a key step toward greater climate resilience for Hong Kong and the wider region.
