Reactivated slope, perched channel, and protective-action capacity: an integrated analysis of the 2025 Matai'an landslide-dam disaster, Taiwan

In the high mountains of eastern Taiwan, the land is alive with movement. The earth here shifts constantly, squeezed by the collision of massive tectonic plates, while heavy rains from typhoons saturate the soil. When a steep slope gives way, it can bury a river valley in tons of rock and mud, creating a natural dam that blocks the water. This is a landslide dam. For a time, the water pools behind this wall of debris, but eventually, the pressure builds, the dam cracks, and the water rushes out in a sudden, violent flood. The danger is not just in the water itself, but in how quickly it arrives. Often, the time between the dam breaking and the flood hitting the towns downstream is shorter than the time it takes for officials to send a warning and for people to move to safety. Scientists study these events to understand the physics of the rock and water, while social scientists study how communities react to warnings. Usually, these two fields of study remain separate, looking at the disaster from different angles. But when a warning fails to save lives, the answer often lies in the space between the geology and the people, where the shape of the land meets the way a town is built and how a government communicates.

On July 21, 2025, a massive landslide occurred in the Matai'an Stream valley, the largest ever recorded in Taiwan. It dumped roughly 308 million cubic meters of rock into the river, creating a dam nearly 200 meters high. For sixty-four days, the water behind this dam rose steadily as a typhoon dumped more rain on the region. On September 23, the water finally topped the wall of rock. Within thirty minutes, the dam burst, sending a torrent of water and debris down the valley. By the time the flood reached the nearby Guangfu Township, it had already claimed nineteen lives. The tragedy was compounded by the fact that the authorities had issued nineteen separate alerts in the days leading up to the disaster. The question researchers asked was simple but profound: why did so many warnings fail to save the people who received them? To find the answer, a team of scientists and social researchers did not look at the warning system or the geology in isolation. Instead, they wove them together, treating the physical landscape, the history of the town, and the instructions given to residents as a single chain of events.

The researchers began by looking at the ground itself, tracing its history back over a century. They found that the slope which failed was not a surprise to those who had studied the land for a long time. Maps drawn in 1924 and 1954 had already marked the area as unstable, showing cracks and contours that indicated a landslide was possible. In the years leading up to the disaster, the land had been shaken by 198 earthquakes of significant strength, and satellite images showed the ground cracking and shifting for eight years prior to the collapse. Despite this long history of visible warning signs, the immediate preparation time given to the residents was measured in minutes, not years. The physical setting of the town played a crucial role in who survived and who did not. The river channel had been narrowed and raised by a century of engineering work, sitting higher than the surrounding flat land like a suspended river. When the flood broke through, it did not spread out evenly. Instead, it was funneled through a specific gap in the embankment, pouring directly onto the lowest, most densely populated streets of the township.

The researchers also examined the history of the people living there. The oldest part of the community, settled by the Amis people before 1904, was located on slightly higher ground. This part of the town was largely spared from the deepest water. However, as the town grew over the following decades, new housing expanded onto the lower, flatter land that had been reclaimed from the river. This newer, lower ground was where the floodwaters concentrated. The study suggests that the original settlers had chosen their location based on generations of experience with the land, but later development pushed people into the most dangerous zones. When the flood arrived, the water depth in these low-lying areas was severe, with sediment deposits reaching over two meters in some places.

The most critical part of the investigation focused on the instructions given to the residents. As the water rose, officials ordered thousands of people to evacuate. For many, the instruction was to perform a "vertical evacuation," meaning they were told to move to an upper floor of their building to escape the rising water. This instruction was given to 5,239 people, representing the largest group of evacuees. However, the researchers found a fatal flaw in how this instruction was applied. The order to move upstairs was issued as a general rule, assuming that the buildings were strong enough and tall enough to keep people safe. The reality on the ground was different. Many of the homes in the hardest-hit areas were single-story structures or buildings made of wood, which offered no upper floor to escape to. Even in two-story concrete buildings, the water and debris were so deep that the second floor might not have been high enough to stay dry.

The study revealed that the government had issued the order to move upstairs without first checking if the specific buildings in the danger zone could actually support that plan. There was no inventory of the buildings that listed which ones were tall enough and strong enough to be safe. The officials knew the addresses of the people who needed to leave, but they did not know the height or construction of the houses at those addresses. Consequently, people in single-story homes were told to go upstairs, a command that was physically impossible for them to follow. The researchers calculated that for the depths of water that actually occurred, a safe refuge would have required a building with at least three stories, yet many of the instructions were given to people in homes that had only one or two.

The failure was not just about the buildings, but also about how the warnings were delivered. The town had a mix of residents, including many elderly people and Indigenous community members who relied less on smartphones and digital apps. The primary warning system used digital alerts, which did not reach these vulnerable groups effectively. While there were other methods like sirens and loudspeakers, these systems were not tested or maintained to the same standard, and the messages they carried were often confusing. The sheer number of alerts—nineteen in a short period—may have also contributed to the problem. When people receive many warnings that turn out to be minor, they can become less responsive to the next one, a phenomenon known as warning fatigue. The researchers noted that the final, life-saving alert was just one in a long sequence of notices, making it harder for residents to distinguish the true emergency from the routine updates.

The study concludes that the tragedy was not an unavoidable act of nature, but the result of a chain of decisions made before the disaster struck. The physical risk had been visible for a century, and the vulnerability of the town was shaped by where people lived and how their homes were built. The warning system failed because it treated the instructions as a generic message rather than a specific plan tailored to the actual conditions of the houses and the people. The researchers argue that for a warning to be effective, the authority giving it must have proof that the action they are asking people to take is actually possible. If a government tells people to move upstairs, they must first know which houses have safe upstairs rooms. If they do not have this information, the instruction is not just unhelpful; it is dangerous.

This analysis offers a new way to think about disaster safety. It suggests that saving lives depends less on how fast a warning can be sent and more on what happens before the disaster begins. By mapping the land, counting the buildings, and understanding the history of the community, officials can identify the gaps between their plans and reality. The researchers propose a simple screening method: before a storm hits, authorities should check every building in a danger zone to see if it meets the height and strength requirements for the expected flood. If a building does not meet the standard, the people inside should be given a different plan, such as being moved to a safe location on the ground. This approach shifts the focus from reacting to the disaster to preparing for it, ensuring that when the water rises, the instructions given to the people are not just words, but a viable path to survival. The Matai'an disaster showed that even with advanced technology and frequent alerts, lives are lost when the physical reality of the town is ignored. The solution lies in connecting the science of the land with the reality of the homes, ensuring that every warning is backed by the evidence that it can be followed.