"Manhole Cover Blasts Open in 12 Seconds... The Reason Behind the 'Gangnam Flood Disaster' [Report]"

Gangnam Flood Disaster Simulated in Andong Lab
Full-scale Replication of Six-lane Roads, Drainage Tunnels, and Stormwater Pump Stations
Utilized as Training Data for AI Flood Forecasting
"Predictions That Took 30 Minutes Now in Under 1 Minute"
Experimental Urban Flood Response in the Era of Climate Crisis

About 10 seconds after the water was released, the surface of the manhole suddenly began to shake. Two seconds later, accompanied by a dull metallic sound, the iron manhole cover shot upwards. This was due to the enormous water pressure pushing up from beneath the manhole.

On May 19, at the River Experiment Center of the Korea Institute of Civil Engineering and Building Technology (KICT) in Andong, North Gyeongsang Province. After the research team blocked the downstream end of the stormwater pipe and released a large volume of water, the "manhole cover ejection" situation—commonly seen in city centers during heavy downpours—was replicated right before their eyes.

Due to sudden heavy rain, the manhole cover, unable to withstand the pressure in the underground drainage system as water surged all at once, popped up, and with the backflow flooding the road, the downtown area quickly turned into a sea of water. Photo by Kim Jonghwa

A powerful column of water and swirling currents burst through the open gap of the manhole in an instant. Among the reporters on site, brief exclamations of surprise erupted.

The "manhole ejection test facility" unveiled that day was constructed to analyze pressure changes and stormwater pipe blockages that occur during urban flooding. The research team was quantitatively measuring under what conditions manhole covers are lifted and at what levels of pressure the risk of pedestrian accidents increases.

"Gangnam Flood Disaster, Recreated in the Laboratory"

The research team then moved to a separately constructed "urban flood pilot test site" to demonstrate a simulation of a Gangnam-style urban flood.

As the experiment began, a massive volume of water poured out simultaneously from four large blue pipes. The water rapidly flowed down an inclined surface designed to mimic a real road, quickly filling a space shaped like a low-lying urban area.

The site was equipped with features such as road markings, curbstones, and storm drains to recreate the look of an actual city. Water was drawn into the stormwater pipes and drainage tunnels, but when the drainage capacity was exceeded, water overflowed onto the road in certain sections.

Sudden heavy rainfall caused rainwater to gather all at once, quickly submerging roads and sidewalks. Photo by Kim Joungho

The "urban flood pilot test site" introduced by KICT that day is not just a simple water tank facility. By applying Seoul Gangnam road design standards, the site recreates a six-lane road, stormwater inlets, pipes, underground stormwater storage facilities, pumping stations, and drainage tunnels, all in full scale to replicate actual urban flood infrastructure.

The test site is connected to four pipes capable of supplying up to 1.2 cubic meters of water per second, and based on its area, it can simulate extreme rainfall scenarios of up to 5,400 millimeters per hour.

Jung Sanghwa, Director of the KICT River Experiment Center, explained, "Urban flooding is not merely a matter of heavy rainfall, but occurs the moment drainage infrastructure is pushed beyond its limits." He added, "The key is to quantitatively analyze under what conditions flooding and manhole ejection occur."

At the site, demonstrations of the drainage tunnel and stormwater pumping station were also conducted. Water pouring through the large blue pipes moved into underground storage facilities beneath the road and was then pumped back out into the external waterway. This structure is a scaled-down version of the actual urban underground drainage system.

"AI to Become the 'Assistant Brain' for Flood Forecasters"

KICT's ultimate goal is not just to recreate flooding scenarios. The core objective is to connect the data obtained from these tests to artificial intelligence (AI) flood forecasting technology.

Currently, the Ministry of Climate and Energy's Flood Control Center is equipped with the AI flood forecasting platform developed by KICT. The number of flood advisory points has expanded from 75 to 223 nationwide, with a significant increase in monitoring for local rivers as well.

Sungkyu Park, President of the Korea Institute of Civil Engineering and Building Technology, is introducing the institute's work. Photo by Jounghwa Kim

Kim Jongmin, a senior researcher at KICT, introduced, "While traditional physics-based prediction models took 30 to 40 minutes to produce results, the AI-based system can provide forecast information for points nationwide within one minute after real-time data input."

KICT is now advancing its AI technology with the goal of "flood prediction six hours in advance." The AI learns in real time using vast hydrological data such as rainfall, water levels, flow rates, and dam discharge volumes to predict flood risks.

In particular, the institute plans to introduce an "AI-physics hybrid model," which combines AI with physics-based numerical analysis, as well as an AI Agent technology to support the entire flood forecasting process.

The researchers referred to AI as "the assistant brain of the forecaster." Their aim is to support not only simple numerical calculations but also comprehensive decision-making regarding the issuance of flood advisories.

"Now Cities Are More at Risk Than Rivers"

KICT predicts that urban flooding will become one of the greatest disasters of the climate crisis era.

While river flooding was the primary flood risk in the past, the institute explains that the rapid increase in "linear rainband" phenomena—where torrential rain concentrates over small areas in a short time—has sharply raised the risk of urban flooding in recent years.

According to KICT data, flood-related casualties dropped to 8 during 2024–2025, when the AI flood forecasting system was fully implemented, compared to 75 in 2023 under the previous system—a reduction of about 89%. However, the research team clarified, "This decrease was not due to AI alone; the expansion of advisories and strengthened government response systems also played a role."

KICT plans to expand its urban flood models to cover semi-basement areas and small to medium-sized cities. Research is also underway on simulations of flooding in the semi-basement neighborhoods around Dorimcheon in Seoul.

Seongyu Park, President of KICT, stated, "In the era of the climate crisis, it is becoming increasingly difficult to respond to floods based solely on past experience," adding, "We will establish a scientific flood response system based on experimental data that accurately replicates real urban environments."

This content was produced with the assistance of AI translation services.