Exclusive: Scientists urge clearer protocol for Taiwan’s barrier lakes

獨家報導：科學家呼籲制定更清晰的台灣堰塞湖管理規程

TAIPEI (Taiwan News) — Two Taiwan engineering professors who examined the Mataian Creek barrier lake flood in Hualien say the country still lacks a clear playbook for landslide-dammed lakes, even when computer models correctly predict where the water will go.

The now-infamous barrier lake in the mountains formed in late July after a large landslide on the upstream section of Mataian Creek. When Typhoon Ragasa brought heavy rain on Sept. 23, the lake overtopped and collapsed, sending a fast-moving mix of water and sediment into Guangfu Township and nearby communities.

The flood killed 19 people and injured more than 150, while several others were listed as missing.

Wang Kuo-lung (王國隆), a civil engineering professor at National Chi Nan University, and Gene Jiing-Yun You (游景雲), a civil engineering professor at National Taiwan University, said the flood exposed weaknesses in how Taiwan prepares for barrier lakes. They point to the timing of risk assessments, the way evacuation orders reached residents, and how key data were shared with researchers.

Wang previously told Taiwan News that Taiwan’s barrier lakes faced “under-monitored landslide risks.” In his follow-up interview, he said one of the main problems in Hualien was the late timing of the scenario simulation that expanded the evacuation zone from hundreds to thousands of residents.

He noted that the lake had been sitting above Guangfu Township for two months, but the full flood scenario was only released on Sept. 19. “The scenario simulation was too late,” he said. “In those last two months, people should have been educated that this was a serious problem.”

Wang said that if authorities had requested and used a detailed simulation earlier, local officials could have briefed residents, planned shelter capacity and run evacuation drills while the weather was still calm. “If they simulated it one month before, then we would have had probably the whole month to educate people, and we could even do drills about this event,” he said.

Peak flows

You’s team at NTU was brought in to assist with the final simulation effort. He said earlier assessments, conducted over a longer period by another group, had suggested a much smaller evacuation zone, even though their calculated peak flows were similar.

“Our estimation for the peak flow was around 1,500 to 1,800 cubic meters per second, and the earlier team had a similar result,” he said. “But when the evacuation area was drawn from that, the number was only around 700 people. From my perspective, that is impossible.”

You said the government asked his team to produce a revised analysis in mid-September and gave them roughly two weeks to do it. Once the updated simulation was submitted and reviewed, he said, “the government only had one day” to turn the new projections into an expanded evacuation on the ground before the typhoon arrived.

Officials ultimately urged close to 8,000 people to leave. Many did evacuate, but not everyone. Both professors said last-minute alerts are not enough for a population that includes many elderly residents and people without easy access to transportation.

Wang said some people stayed in their homes because they lacked cars or the ability to move elderly relatives to higher ground. Others did not receive the warnings clearly. He said the region’s geography and age profile made evacuation more complicated than it appeared on paper.

Government broadcasts, mobile phone alerts and police loudspeakers were used. However, You said, mountainous terrain and the digital divide created gaps in who actually received and acted on those messages.

Modeling disaster

“In some areas the loudspeakers could not reach people, and some seniors do not use smartphones,” he said. He added that police and firefighters had limited manpower for door-to-door checks in the short time available.

You said he does not believe residents ignored evacuation orders. He said most households that received clear information moved out quickly, but the reasons some residents remained in place are still not fully understood.

Wang said a barrier lake that persists for weeks or months requires a different kind of public preparation than an ordinary river flood. He compared it to Taiwan’s regular training for communities living near debris-flow channels, where drills and briefings are conducted each year.

“In this case there was no drill and no education,” he said. “It is not a traditional floodplain, so people did not pay attention. He added that when a barrier lake exists for that long, people need to know how serious it can be.

Both professors also pointed to ways Taiwan could improve monitoring at unstable natural dams. Wang said vibration sensors installed on or near the dam can detect signs of internal instability hours or days before collapse, and ground-based SAR radar or optical systems can measure deformation even at night or in cloud cover.

“These tools can give early warning,” he said. “But evacuation needs a process. You need education, training and drills to save people’s lives.”

Not complete failure

You said the accuracy of outburst-flood simulations depends heavily on high-resolution elevation data. He noted that the hydrological digital elevation model, or hydro-DEM, for Mataian Creek was not publicly accessible and had to be requested during the crisis.

Modelers in Taiwan can usually access standard elevation datasets, he said, but hydro-DEMs involve more detailed processing of river channels and are often limited to government agencies. He said he believes broader access would help scientific teams run cross-checks and refine assumptions earlier.

“If more groups could access the data before the emergency, the modeling would be more rigorous,” he said. He added that uncertainty is unavoidable in such events, and that the good alignment between the final NTU simulation and the real flood was partly “a little lucky.”

Neither researcher described the response as a complete failure, but both said barrier lakes should be treated as a distinct hazard category. They said more transparent data, earlier scenario work and clear local protocols would help integrate scientific analysis with community-level action.

Wang said the Mataian case should become a reference point for future planning, not just a tragic memory. He called it “a textbook event,” and said it should be used in that way.

He said that if barrier lakes form in the future and data are opened earlier to research teams, more specialists will be able to contribute to risk assessments and evacuation planning.

“If they can open data earlier, more scientists can join the work,” he said. “If they can be involved in the prediction or scenario simulation, then we can help save people’s lives in the future.”