Due to climate change, unpredictable heavy rain is becoming more frequent, leading to increased large-scale flood damage across the country. In this context, domestic researchers have utilized artificial intelligence (AI) to predict flood risks by region and have created a nationwide "flood risk map."
Research by Professor Jang Jong-hoon of Pohang University of Science and Technology POSTECH and Professor Jung Young-hoon of Kyungpook National University was published on the 28th in the environmental science journal "Journal of Environmental Management."
Flood damage is becoming increasingly serious due to climate change and rapid urbanization. In particular, the increase of concrete roads and buildings that do not allow rainwater to seep in is amplifying damage, even with the same amount of rain. While the "analytical hierarchy process (AHP)" was primarily used to predict flood risk, this method requires a lot of time and expense and is difficult to express the reliability of the predictions numerically.
The research team addressed this issue using AI. They first analyzed flood damage data recorded by the Ministry of the Interior and Safety for the past 20 years (2002-2021) by city and county. Based on this, they delineated four key elements that determine flood risk: "hazard (how much rain falls)," "exposure (population and infrastructure exposed to risk)," "vulnerability (degree of susceptibility to damage)," and "coping capacity (how well one can respond)," and trained these aspects with AI.
As a result, among several AI models, the "XGBoost" and "Random Forest" models predicted flood damage with over 77% accuracy. Interestingly, the two models identified different factors as the most significant risks. XGBoost analyzed the "ratio of impenetrable surfaces (impervious surface ratio)" as the biggest risk factor, while Random Forest pinpointed "river area."
Notably, both AI models assessed large cities such as Seoul and Incheon as "high-risk flood areas." This indicates that these areas, with high population density, extensive concrete pavement, and concentrated buildings and infrastructure near rivers, are more vulnerable to damage.
Subsequently, the research team proposed practical solutions. Given that the "impervious surface ratio" and "river area" were identified as major risk factors through AI analysis, they emphasized the need for nature-friendly urban development policies, such as ensuring green spaces that allow rainwater to naturally absorb into the ground and restricting development around rivers, to reduce flood damage.
Professor Jang Jong-hoon noted, "AI cannot perfectly assess every situation, so it should still be used alongside expert judgment, and it will produce more accurate flood inundation maps." Professor Jung Young-hoon stated, "By utilizing big data related to floods and AI, we will create region-specific flood inundation risk maps that will provide important data for future localized flood and inundation response measures."
References
Journal of Environmental Management (2025), DOI: https://doi.org/10.1016/j.jenvman.2025.125640