WeatherPlus unveils technology to forecast flood peak 10 hours in advance

Dr. Ha Ngoc Tuan, representing the Kyushu-WeatherPlus consortium, stated that the forecasting model relies on real-time data from over 700 rainfall, flow, and water level monitoring stations. Data from these stations are continuously transmitted to the processing center and then integrated with flow simulations to develop reservoir operation scenarios during heavy rainfall events.

According to the development team, the goal of the technology is to convert rainfall information into forecasts of inflow volume, water level rise rate, and potential impacts on downstream areas.

During recent extreme rainfall events, these real-time simulations have been used to estimate when flood peaks might occur. One example presented at the forum “Application of Science and Technology in Disaster Forecasting and Early Warning” on the afternoon of November 25 was the flood in the Ba Ha River basin.

The basin, covering more than 11,000 km², recorded an average rainfall of 250-300 mm over several days, equivalent to approximately 2.85-2.87 billion m³ of water. This volume is nearly double the total capacity of major reservoirs such as Thac Ba (around 1.25 billion m³) or Ban Ve (1.4 billion m³) during strong floods caused by storms Yagi (2024) or Wipha (2025).

With such rapid and large-scale increases in water, real-time simulations help operators anticipate how reservoir levels will rise and when to open spillways, according to Dr. Tuan.

The operation principle is modeled based on the natural delay between upstream rainfall and downstream water levels. Observations at the Ba Ha River show a lag of about 9-10 hours from the onset of rain to high water at populated areas, with a minimum of around 5 hours even under adverse conditions.

The developers note that if forecasts can “see ahead” during this time window, disaster response agencies will have a solid basis to make earlier evacuation decisions and avoid reactive measures.

However, the team also acknowledged that forecast errors are inevitable. During the recent flood on the Ba Ha River, some international models overestimated rainfall in the first event, while underestimating it in the subsequent downpour. The domestic operational model used by Weatherplus showed an error of over 15% during the second heavy rainfall. Nevertheless, the team believes that being able to identify rainfall amounts of hundreds of millimeters over a basin spanning tens of thousands of square kilometers is still "reliable enough" to assess risk and prepare safety measures.

“The prerequisite for accurate system operation is data quality. Rain gauges are continuously checked, erroneous data must be removed, and all input information must be standardized before being used in simulations,” said Dr. Ha Ngoc Tuan. He added that without complete or clean data, no model can produce reliable results.

Examples of reservoir operations were cited, such as forecasting a sharp increase in flow during Typhoon Wipha, which allowed the Hua Na hydropower plant to prepare regulation plans more than a day in advance, or simulations during Typhoon Kajiki showing that water levels at Bai Thuong Hydropower would remain within safe limits if operated according to pre-calculated scenarios.

Although recently disclosed by the company, these examples demonstrate how science and technology can transform rainfall data into actionable operational decisions.

In particular, Dr. Tuan's team demonstrated the potential to establish a water-level-based warning system at a reference station. Specifically, when water reaches a certain level, authorities can prepare for the evacuation of vulnerable groups. If it exceeds another threshold, all residents in the danger zone must evacuate. According to Weatherplus, this approach is more intuitive than the traditional alert levels I, II, or III, which can be difficult for people to visualize.

The team expressed the hope that, once fully developed, the HNT technology could be widely applied, including the design of simple warning tools for local authorities. They emphasized that identifying the peak flood 5-10 hours in advance is crucial for minimizing casualties and reducing damage downstream. After three years of testing on major reservoirs, Dr. Tuan and his colleagues reaffirmed the importance of reliable observational data and forecasts. “The real question is whether we are willing to invest in data, in people, and in science and technology,” he concluded.