'Magic eye' for disaster forecasting and early warning

The early warning system helps to "see" the flood peak 5-10 hours in advance.

On the afternoon of November 25, the Vietnam Agriculture and Nature Newspaper, in coordination with the Vietnam Disaster and Dyke Management Authority (Ministry of Agriculture and Environment), the Disaster Prevention Community Fund, and the Bac Ninh Department of Agriculture and Environment, organized the Forum "Application of Science and Technology in Disaster Forecasting and Early Warning."

According to statistics, during the 2006-2025 period, natural disasters caused an average of 314 deaths and missing persons annually, with economic losses equivalent to 1-1.5% of GDP. In 2025, extreme heavy rainfall caused floods exceeding historical levels on more than 13 river routes and severe flooding in many urban areas from the Northern Delta to the Central Highlands, creating a situation of "typhoons stacking upon typhoons, floods stacking upon floods."

In this context, Dr. Cao Duc Phat, former Minister of Agriculture and Rural Development (MARD) and Chairman of the Management Council of the Disaster Prevention Community Fund, emphasized that in disaster response, monitoring, tracking, forecasting, and early warning play a key role. The strong development of science and technology, especially digital technology and AI, is opening up new opportunities to enhance forecasting effectiveness. In addition to satellite data, various automated devices for measuring rain, radar, sensors for measuring temperature, wind, water levels, flow rate, or earth movement are being deployed, allowing for faster information integration, more accurate modeling tailored to each locality, and timely transmission to the public.

According to Dr. Ha Ngoc Tuan, representative of the Kyushu - WeatherPlus Consortium, the HNT Reservoir Operation Support System, developed with Kyushu Electric Power Co., Inc. (Japan), allows for real-time forecasting of rain, flow, and flood discharge simulation. The system provides operators with 5-10 hours of "golden time" to protect structures and evacuate residents during extreme floods.

HNT is one of the few platforms in Vietnam that synchronously applies AI, Big Data, numerical models, and real-time computation in reservoir management. Data is collected from over 700 rain, flow, and water level monitoring stations, transmitted continuously via the telecommunications network. Based on this, quantitative rainfall forecasting models, hydrological - hydraulic models, and single- and multi-reservoir computation are "integrated" into a digital brain, analyzing and displaying real-time information, helping operators foresee multiple flood scenarios to make quick decisions when flow changes.

The key difference of HNT is that it transforms rain into operational information, from forecasting inflow into the reservoir, flood discharge scenarios, to reservoir water level and downstream impact. During Typhoon Wipha, the HNT bulletin was sent to the Hua Na Hydropower Plant approximately 30 hours before the flood peak, building a scenario for a peak flow of 7,300 cubic meters per second, helping to optimize reservoir capacity operation, ensuring safety for the structure and downstream areas. A similar situation occurred at Bai Thuong Hydropower Plant during Typhoon Kajiki, where HNT forecasting helped determine that the water level would not exceed the dangerous threshold, allowing the plant to operate safely.

Dr. Tuan analyzed the flood on the Ba Ha River when the average rainfall was 250−300 mm, which was double the total inflow to the Thac Ba and Ban Ve reservoirs during Typhoons Yagi (2024) and Wipha (2025), respectively. The expert emphasized that a limitation of forecasting models is that global and regional models still have an error of about 15%, with a tendency to over- or under-forecast compared to reality. However, the overall information clearly shows the extreme event, which is sufficient for response preparations.

The greatest value of HNT is transforming rainfall figures (from flow rate) into specific warnings and actions. The analysis of the Ba Ha River shows that from the moment rain falls until the flood peaks downstream, it takes an average of 9-10 hours, meaning there are still about 5 hours of "golden time" to evacuate residents. The warnings can be specific, such as the elderly and children preparing to evacuate when the water level approaches Threshold A; and the entire population being required to evacuate before Threshold B.

With rainfall data across the basin, through analysis and correlation with data from on-site flood measurement solutions like those from Watec Company, disaster prevention agencies can forecast flooding without needing to deploy complex and expensive models like real-time flood maps.

Early warning for landslides and flooding in key areas

Sharing the solution for landslide detection and building safe residential areas based on geological data, machine learning technology, and real-time warnings, Prof. Dr. Do Minh Duc (University of Science - Vietnam National University, Hanoi) stated that the model has been tested in Mu Cang Chai town, an area frequently affected by landslides and debris flows. The survey showed that many large slip points appeared near residential areas, including locations where slopes were cut during house construction, causing terrain instability and obstructing water flow.

From this reality, the research team developed a set of risk assessment criteria, created a list of households located in high-risk areas, and proposed a model for safe residential areas based on three principles: compliance with land and construction regulations, proactive prevention, and zero human loss. The risk zoning map was built using the Random Forest machine learning model, allowing for high-detail identification of areas prone to landslides.

A key highlight of the solution is the "Viet Nam Landslide" application, which integrates rainfall maps, risk maps, and hourly warnings. Rain gauges and warning sirens have been installed at various points within the residential area to support timely evacuation. Prof. Dr. Do Minh Duc emphasized three key lessons. These are early detection - early warning - early action; the right people - the right job - at the right time; and three levels of information from receipt to transformation into action. These are considered the foundation for localities to be more proactive against extreme weather phenomena.

In the context of the increasingly urgent demand for early warning, Mr. Le Viet Xe, Deputy Director of WATEC Company, stated that the VFASS automated flood monitoring system has been researched and deployed. The device uses a radar sensor to measure inundation depth, transmits data continuously via 3G/4G networks, is powered by solar energy, and automatically issues warnings when the water level exceeds a threshold.

The data is processed on a cloud computing platform, allowing for centralized management and sharing with authorities and residents via web and mobile applications. VFASS can connect with rain gauges, water level meters, and flood warning towers to issue multi-channel warnings in real time. To date, the system has been deployed with approximately 140 flood warning towers and monitoring stations nationwide, and has also begun being exported abroad. Mr. Xe affirmed that automated monitoring solutions will play an important role in forecasting, developing response scenarios, and minimizing damage caused by flooding.

Over the past 17 years, the Disaster Prevention Community Fund has mobilized social resources to install nearly 1,000 automatic rain gauge and flood warning stations; it is currently continuing to coordinate support for salinity measurement equipment, landslide warnings, the construction of digital warning systems, and grassroots disaster prevention assault teams.

In early warning efforts, from 2016 to 2025, the Fund sponsored the installation of rain gauges, flood warning towers, and landslide early warning systems in many localities, contributing to enhanced response capabilities for both the community and the local government.