AI-powered landslip warning system with 90 percent accuracy to launch in 2026
Hong Kong will roll out a fifth-generation, AI-driven landslip warning system citywide in 2026, aiming to deliver earlier, more precise alerts and bolster emergency response. Built on decades of geotechnical data and machine learning, the platform has demonstrated over 90 percent accuracy in tests—up from roughly 70 percent with the current system.
Briefing reporters on Tuesday, the Geotechnical Engineering Office (GEO) said the new system combines artificial intelligence with big data to generate tailor-made prediction models. By learning from hundreds of past rainstorms and thousands of slope records, the system is designed to pinpoint landslide risk at finer spatial and temporal scales, improving public warnings and resource deployment during severe weather.
How the model was trained
The AI has been trained on more than 380 rainfall events associated with documented landslides between 1996 and 2023. Its training corpus spans:
- Data on around 60,000 man-made slopes
- Approximately 22 million slope-related data points
- About 2,700 reported landslip cases
According to GEO head Raymond Cheung Wai-man, the system’s continuous-learning design means prediction accuracy can keep improving as more data are collected. The GEO plans to track advances in machine learning and adopt suitable algorithms to further enhance assessments of landslide risk.
Trials point to higher accuracy
Senior geotechnical engineer Edward Chu Kei-hong said the GEO ran real-time trials during the rainy season starting in April and May this year, focusing on dynamic risk estimation as conditions evolved. He described the early results as encouraging and noted that optimization and fine-tuning are ongoing ahead of full deployment.
What’s new in the fifth generation
Since Hong Kong introduced the world’s first regional landslide warning system in 1977, the platform has steadily incorporated new data sources and analytical techniques. The latest generation expands the range of rainfall and slope features considered, enabling:
- More granular, tailor-made prediction models based on local terrain and slope characteristics
- Sharper detection of elevated risk windows during intense rain
- Improved targeting of alerts, including territory-wide warnings and, when needed, regional special advisories
Why it matters
Many of Hong Kong’s neighborhoods sit near steep slopes, heightening vulnerability to landslides during heavy downpours. The city’s risk was seared into memory by the deadly June 18, 1972 landslides at Sau Mau Ping and Po Shan Road, which destroyed buildings and claimed 138 lives—events that helped catalyze the development of the GEO’s warning system.
Cheung noted that more than 90 percent of fatal landslide accidents have occurred during periods when landslip warnings were already in force, underscoring how quickly hazards can escalate and the importance of clear, timely communication. Beyond accuracy improvements, the GEO is prioritizing better information dissemination and will issue regional advisories when specific districts face heightened risk.
Timeline and next steps
The AI-powered system is slated for full rollout in 2026. Until then, the GEO will continue live trials, expand training datasets, and refine models to improve reliability across varied storm patterns. The office also plans closer coordination with emergency responders to ensure alerts translate into faster on-the-ground action.
For residents, the shift means more precise warnings, stronger guidance during severe rain, and a system that can learn from each event—raising the bar for landslide resilience across Hong Kong.
